The COVID-19 pandemic began five years ago, seemingly all at once. Treatments quickly became ... [+] available, primarily because of federally funded basic research. (Photo by John Paraskevas/Newsday RM vis Getty Images)Newsday via Getty ImagesFive years ago, the COVID-19 pandemic seemed to begin all at once. It officially started with a declaration by the World Health Organization on March 11, 2020. Within days, schools and restaurants closed. The death toll continued to rise. Hospitals were overwhelmed. Our world had changed.But medical advancements seemed to occur all at once, too. Just seven months after the pandemic declaration, an antiviral that helped reduce disease severity and death became available. Two months after that, highly effective and safe mRNA vaccines became available. These advancements, however, didnt actually occur all at once. Their rapid development was possible because of basic research, often federally funded, that had been ongoing for decades.With funding cuts proposed by President Trump, this basic research is in jeopardy. If another pandemic occurs, we may not be able to respond with warp speed.To more fully explore the importance of federally funded basic research, lets look at the remdesivir story. The FDA approved this antiviral compound for the treatment of COVID-19 on October 22, 2020. Dr. Mark Denison, Director of the Division of Pediatric Infectious Diseases at Vanderbilt University Medical Center, contributed to the development of remdesivir and another COVID-19 drug, molnupiravir. But he didnt start out with that goal in mind.Denison has been studying coronaviruses for four decades. In a recent conversation, he noted that his work on remdesivir began with a much different goal, investigating the role of the coronavirus ExoN protein. Described as weird by Dr. Denison, ExoN functions as a proofreading enzyme. It allows the virus to fix mistakes that occur during replication, a property not shared by many other viruses.Dr. Denison wanted to learn more about the how this enzyme contributed to coronavirus replication. Ive researched coronaviruses for 40 years. Im interested in how they work, he said. In retrospect, his approach seems straight-forward. We wanted to find a compound that inhibited this enzyme. Then we showed that this compound inhibited viral replication in cell culture and infection in animals. Finally, we showed that it was effective in humans. He noted that the same playbook was used during the development of molnupiravir.This work wasnt initiated in response to the COVID-19 pandemic. The studies were conducted primarily to investigate interesting questions about coronavirus biology. As Dr. Denison remarked, Finding antivirals wasnt my career choice.The development of the mRNA vaccines presents another example of the value of basic research. Drs. Katalin Karik and Drew Weissman received the 2023 Nobel Prize in Physiology or Medicine for their seminal work leading to the development of the Covid-19 vaccines. As noted in the official Nobel Prize press release, The vaccines have saved millions of livesallowing societies to open and return to normal conditions.The award, however, was not for the actual development of the vaccines. Karik and Weissman were recognized for basic research into our immunological response to mRNA and an atypical component of RNA called pseudouridine. They were recognized for work done over 15 years ago, long before the pandemic began. And the story certainly didnt start with them. Work by numerous researchers on topics like the synthesis of RNA and the delivery of RNA into cells preceded their work, thereby setting the stage.Science is an iterative process. And often, as shown by the work of Drs. Denison, Karik and Weissman, the end result isnt even a consideration when the work begins.The basic research that eventually led to the development of remdesivir, molnupiravir, the mRNA vaccines and other COVID-19 treatments largely was funded by the National Institutes of Health. With a budget of nearly $48 billion, the institutes support over 300,000 researchers at over 2,500 institutions, according to the NIH website.The results of this funding are remarkable. Over 170 scientists employed or funded by the NIH are Nobel Prize recipients, including Karik and Weissman. Countless other researchers use this federal funding to make small but important scientific advances. Most likely, many of them are motivated largely by curiosity. They investigate myriad biological systems because, as Dr. Denison noted, they are interested in how they work.Recent proposals to cut funding to the NIH could derail this basic research. In February, the Trump administration proposed a drastic cut in the indirect costs associated with its grants. These overhead costs help institutions maintain important services, like animal care facilities. The loss of these funds could be devastating and negatively impact the basic research occurring at universities and medical schools throughout the country.Today, over 7 million people globally and 1.2 million people in the United States have died as a result of the COVID-19 pandemic. Seventeen million adults in the U.S. may have long COVID, according to KFF. Yes, our world has fundamentally changed. But it could have been worse. Next time, it might be.

In brief: With so much of the world now online and more people considering themselves tech-savvy, one might imagine that scammers are struggling to con people out of money. The reality is that Americans lost more to fraud than ever before in 2024 - $12.5 billion. According to the FTC, the $12.5 billion Americans lost in 2024 to scams was up 25% compared to the $10 billion losses a year earlier.Surprisingly, the actual number of people who reported being the victim of a fraud, 2.6 million, didn't increase in 2024. What did increase was the percentage of people who lost money in these scams up from 27% in 2023 to 38% last year. It seems scammers are getting better at convincing people to part with their cash.The highest losses came from investment scams. A massive $5.7 billion was lost through this type of fraud. The median loss for victims was over $9,000, more than any other scam category.Often known as pig butchering, a reference to fattening the animal up before slaughter, these investment frauds involve grooming victims over time, winning their trust before convincing them to make large investments. Once someone gives the scammer their money, the criminal cuts off all communications.The second most-successful type of fraud, and the one that was most commonly reported, was the imposter scam. The FTC writes that $2.95 billion was lost by victims of these crimes, which involve a scammer pretending to be a government official, business representative, family member, or a romantic partner and convincing someone to hand over money. Losses to government imposter scams in particular increased by $171 million from 2023 to a total of $789 million in 2024.It's a stereotype that the elderly are most likely to fall victim to scams, but younger people reported losing money to fraud more often than those over 70 44% of all reports were filed by those between 20 and 29. This could be due to a greater number of young people being online. // Related StoriesElsewhere, losses to business and job opportunity scams surged, from $250 million to $750.6 million.For the second year in a row, the most common method that scammers used to make first contact with victims was email. This was followed by phone calls and then text messages. However, while people lost $1.9 billion to scams that began with contact using these more traditional methods, People lost over $3 billion to scams that started online.It's worth remembering that the majority of scams aren't reported, usually because the victim feels too embarrassed, so the actual figures are likely much higher.

What just happened? In a court filing earlier this month, U.S. federal agents confirmed that a series of high-profile cyberheists, including a $150 million cryptocurrency theft, are linked to the 2022 breach of password manager service LastPass. The heists involved cracking master passwords stolen from LastPass, which allowed thieves to access sensitive information, including cryptocurrency seed phrases stored in the "Secure Notes" section of victims' accounts, according to KrebsonSecurity, which has been tracking these incidents since September 2023. The $150 million heist, which occurred on January 30, 2024, is believed to have targeted Chris Larsen, co-founder of the cryptocurrency platform Ripple, according to blockchain security researcher ZachXBT. Federal prosecutors in northern California have seized approximately $24 million in cryptocurrencies related to this theft.According to the seizure document, the U.S. Secret Service and the FBI believe the attackers used stolen data from LastPass to access victims' accounts without authorization. This pattern is consistent with similar six-figure crypto heists, where victims had stored their cryptocurrency seed phrases in LastPass before the 2022 breaches.Krebs says that security researchers Nick Bax and Taylor Monahan have been working with dozens of victims and found none experienced typical precursor attacks, such as email or mobile phone account compromises, or SIM-swapping attacks. Instead, all victims had stored their cryptocurrency seed phrases in LastPass's "Secure Notes" before the breaches. The heists followed a similar pattern of rapidly moving stolen funds to numerous drop accounts scattered across various cryptocurrency exchanges.The breach of LastPass in 2022 involved two significant incidents. Initially, on August 25, 2022, LastPass CEO Karim Toubba announced that the company had detected unusual activity in its software development environment, resulting in the theft of some source code and proprietary technical information.However, on September 15, 2022, LastPass stated that the investigation found no access to customer data or password vaults. This assessment changed on November 30, 2022, when LastPass disclosed that criminal hackers had compromised encrypted copies of some password vaults and other personal information using data stolen in the August breach. // Related StoriesThis breach would have given thieves offline access to encrypted password vaults, allowing them to attempt to crack weaker master passwords using powerful systems capable of millions of guesses per second. Many victims had chosen master passwords with relatively low complexity and were among LastPass's oldest customers.Legacy users were more likely to have master passwords protected with fewer iterations the number of times a password is run through the company's encryption routines. The more iterations, the longer it takes an offline attacker to crack the master password. Over the years, LastPass increased the number of iterations for new users, requiring longer and more complex master passwords. However, researchers found that many older customers were not upgraded to these newer security standards.Despite these findings, LastPass maintains no definitive proof linking the cyberheists to their breaches. The company says it has been cooperating with law enforcement and investing in enhanced security measures.However, researchers have expressed concern that LastPass has not adequately alerted its customers about the potential risks, particularly sensitive information stored in "Secure Notes." They argue that more proactive measures could have prevented millions of dollars in thefts.Bax noted that after issuing the initial warning, he hoped users would migrate their funds to new cryptocurrency wallets. However, the continued thefts show how much more needs to be done.LastPass could have encouraged users to rotate their credentials and prevented further thefts but instead chose to deny the risks and blame the victims, Monahan said. The situation remains critical, with recent reports of additional thefts in December.

Love crossword puzzles but dont have all day to sit and solve a full-sized puzzle in your daily newspaper? Thats what The Mini is for!A bite-sized version of the New York Times well-known crossword puzzle, The Mini is a quick and easy way to test your crossword skills daily in a lot less time (the average puzzle takes most players just over a minute to solve). While The Mini is smaller and simpler than a normal crossword, it isnt always easy. Tripping up on one clue can be the difference between a personal best completion time and an embarrassing solve attempt.Recommended VideosJust like ourWordle hints and Connections hints, were here to help with The Mini today if youre stuck and need a little help.Please enable Javascript to view this contentBelow are the answers for the NYT Mini crossword today.New York TimesAcrossFormat for many e-signed documents PDFAspirations GOALSThis thesaurus is awful. Not only is it awful, its also ___! (joke) AWFULSouthpaw, by another name LEFTYDeli bread variety RYEDownWord after flower or fire POWERLooney Tunes character with the catchphrase Youre dethpicable! DAFFYWhat you might blow in, when youre blowin in the wind? FLUTEActress Gadot GALAdjective for a fox SLYEditors Recommendations

Table of ContentsTable of ContentsHow to play StrandsHint for todays Strands puzzleTodays Strand answersStrands is a brand new daily puzzle from the New York Times. A trickier take on the classic word search, youll need a keen eye to solve this puzzle.Like Wordle, Connections, and the Mini Crossword, Strands can be a bit difficult to solve some days. Theres no shame in needing a little help from time to time. If youre stuck and need to know the answers to todays Strands puzzle, check out the solved puzzle below.Recommended VideosHow to play StrandsYou start every Strands puzzle with the goal of finding the theme words hidden in the grid of letters. Manipulate letters by dragging or tapping to craft words; double-tap the final letter to confirm. If you find the correct word, the letters will be highlighted blue and will no longer be selectable.RelatedIf you find a word that isnt a theme word, it still helps! For every three non-theme words you find that are at least four letters long, youll get a hint the letters of one of the theme words will be revealed and youll just have to unscramble it.Every single letter on the grid is used to spell out the theme words and there is no overlap. Every letter will be used once, and only once.Each puzzle contains one spangram, a special theme word (or words) that describe the puzzles theme and touches two opposite sides of the board. When you find the spangram, it will be highlighted yellow.The goal should be to complete the puzzle quickly without using too many hints.Todays theme is Whats the magic word?Heres a hint that might help you: words of power.Todays Strand answersNYTTodays spanagramWell start by giving you the spangram, which might help you figure out the theme and solve the rest of the puzzle on your own:CASTINGSPELLSTodays Strands answersPRESTOALAKAZAMTADASHAZAMABRACADABRAEditors Recommendations

Optical Character Recognition Why extracting data from PDFs is still a nightmare for data experts Countless digital documents hold valuable info, and the AI industry is attempting to set it free. Benj Edwards Mar 11, 2025 7:15 am | 17 Credit: Vertigo3d via Getty Images Credit: Vertigo3d via Getty Images Story textSizeSmallStandardLargeWidth *StandardWideLinksStandardOrange* Subscribers only Learn moreFor years, businesses, governments, and researchers have struggled with a persistent problem: How to extract usable data from Portable Document Format (PDF) files. These digital documents serve as containers for everything from scientific research to government records, but their rigid formats often trap the data inside, making it difficult for machines to read and analyze."Part of the problem is that PDFs are a creature of a time when print layout was a big influence on publishing software, and PDFs are more of a 'print' product than a digital one," Derek Willis, a lecturer in Data and Computational Journalism at the University of Maryland, wrote in an email to Ars Technica. "The main issue is that many PDFs are simply pictures of information, which means you need Optical Character Recognition software to turn those pictures into data, especially when the original is old or includes handwriting."Computational journalism is a field where traditional reporting techniques merge with data analysis, coding, and algorithmic thinking to uncover stories that might otherwise remain hidden in large datasets, which makes unlocking that data a particular interest for Willis.The PDF challenge also represents a significant bottleneck in the world of data analysis and machine learning at large. According to several studies, approximately 8090 percent of the world's organizational data is stored as unstructured data in documents, much of it locked away in formats that resist easy extraction. The problem worsens with two-column layouts, tables, charts, and scanned documents with poor image quality.The inability to reliably extract data from PDFs affects numerous sectors but hits hardest in areas that rely heavily on documentation and legacy records, including digitizing scientific research, preserving historical documents, streamlining customer service, and making technical literature more accessible to AI systems."It is a very real problem for almost anything published more than 20 years ago and in particular for government records," Willis says. "That impacts not just the operation of public agencies like the courts, police, and social services but also journalists, who rely on those records for stories. It also forces some industries that depend on information, like insurance and banking, to invest time and resources in converting PDFs into data."A very brief history of OCRTraditional optical character recognition (OCR) technology, which converts images of text into machine-readable text, has been around since the 1970s. Inventor Ray Kurzweil pioneered the commercial development of OCR systems, including the Kurzweil Reading Machine for the blind in 1976, which relied on pattern-matching algorithms to identify characters from pixel arrangements.These traditional OCR systems typically work by identifying patterns of light and dark pixels in images, matching them to known character shapes, and outputting the recognized text. While effective for clear, straightforward documents, these pattern-matching systems, a form of AI themselves, often falter when faced with unusual fonts, multiple columns, tables, or poor-quality scans.Traditional OCR persists in many workflows precisely because its limitations are well-understoodit makes predictable errors that can be identified and corrected, offering a reliability that sometimes outweighs the theoretical advantages of newer AI-based solutions. But now that transformer-based large language models (LLMs) are getting the lion's share of funding dollars, companies are increasingly turning to them for a new approach to reading documents.The rise of AI language models in OCRUnlike traditional OCR methods that follow a rigid sequence of identifying characters based on pixel patterns, multimodal LLMs that can read documents are trained on text and images that have been translated into chunks of data called tokens and fed into large neural networks. Vision-capable LLMs from companies like OpenAI, Google, and Meta analyze documents by recognizing relationships between visual elements and understanding contextual cues.The "visual" image-based method is how ChatGPT reads a PDF file, for example, if you upload it through the AI assistant interface. It's a fundamentally different approach than standard OCR that allows them to potentially process documents more holistically, considering both visual layouts and text content simultaneously.And as it turns out, some LLMs from certain vendors are better at this task than others."The LLMs that do well on these tasks tend to behave in ways that are more consistent with how I would do it manually," Willis said. He noted that some traditional OCR methods are quite good, particularly Amazons Textract, but that "they also are bound by the rules of their software and limitations on how much text they can refer to when attempting to recognize an unusual pattern." Willis added, "With LLMs, I think you trade that for an expanded context that seems to help them make better predictions about whether a digit is a three or an eight, for example."This context-based approach enables these models to better handle complex layouts, interpret tables, and distinguish between document elements like headers, captions, and body textall tasks that traditional OCR solutions struggle with."[LLMs] arent perfect and sometimes require significant intervention to do the job well, but the fact that you can adjust them at all [with custom prompts] is a big advantage," Willis said.New attempts at LLM-based OCRAs the demand for better document-processing solutions grows, new AI players are entering the market with specialized offerings. One such recent entrant has caught the attention of document-processing specialists in particular.Mistral, a French AI company known for its smaller LLMs, recently entered the LLM-powered optical reader space with Mistral OCR, a specialized API designed for document processing. According to Mistral's materials, their system aims to extract text and images from documents with complex layouts by using its language model capabilities to process document elements. Credit: Kirillm via Getty Images However, these promotional claims don't always match real-world performance, according to recent tests. "I'm typically a pretty big fan of the Mistral models, but the new OCR-specific one they released last week really performed poorly," Willis noted."A colleague sent this PDF and asked if I could help him parse the table it contained," says Willis. "It's an old document with a table that has some complex layout elements. The new [Mistral] OCR-specific model really performed poorly, repeating the names of cities and botching a lot of the numbers."AI app developer Alexander Doria also recently pointed out on X a flaw with Mistral OCR's ability to understand handwriting, writing, "Unfortunately Mistral-OCR has still the usual VLM curse: with challenging manuscripts, it hallucinates completely."According to Willis, Google currently leads the field in AI models that can read documents: "Right now, for me the clear leader is Google's Gemini 2.0 Flash Pro Experimental. It handled the PDF that Mistral did not with a tiny number of mistakes, and I've run multiple messy PDFs through it with success, including those with handwritten content."Gemini's performance stems largely from its ability to process expansive documents (in a type of short-term memory called a "context window"), which Willis specifically notes as a key advantage: "The size of its context window also helps, since I can upload large documents and work through them in parts." This capability, combined with more robust handling of handwritten content, apparently gives Google's model a practical edge over competitors in real-world document-processing tasks for now.The drawbacks of LLM-based OCRDespite their promise, LLMs introduce several new problems to document processing. Among them, they can introduce confabulations or hallucinations (plausible-sounding but incorrect information), accidentally follow instructions in the text (thinking they are part of a user prompt), or just generally misinterpret the data."The biggest [drawback] is that they are probabilistic prediction machines and will get it wrong in ways that aren't just 'that's the wrong word'," Willis explains. "LLMs will sometimes skip a line in larger documents where the layout repeats itself, I've found, where OCR isn't likely to do that."AI researcher and data journalist Simon Willison identified several critical concerns of using LLMs for OCR in a conversation with Ars Technica. "I still think the biggest challenge is the risk of accidental instruction following," he says, always wary of prompt injections (in this case accidental) that might feed nefarious or contradictory instructions to a LLM."That and the fact that table interpretation mistakes can be catastrophic," Willison adds. "In the past I've had lots of cases where a vision LLM has matched up the wrong line of data with the wrong heading, which results in absolute junk that looks correct. Also that thing where sometimes if text is illegible a model might just invent the text."These issues become particularly troublesome when processing financial statements, legal documents, or medical records, where a mistake might put someone's life in danger. The reliability problems mean these tools often require careful human oversight, limiting their value for fully automated data extraction.The path forwardEven in our seemingly advanced age of AI, there is still no perfect OCR solution. The race to unlock data from PDFs continues, with companies like Google now offering context-aware generative AI products. Some of the motivation for unlocking PDFs among AI companies, as Willis observes, doubtless involves potential training data acquisition: "I think Mistral's announcement is pretty clear evidence that documentsnot just PDFsare a big part of their strategy, exactly because it will likely provide additional training data."Whether it benefits AI companies with training data or historians analyzing a historical census, as these technologies improve, they may unlock repositories of knowledge currently trapped in digital formats designed primarily for human consumption. That could lead to a new golden age of data analysisor a field day for hard-to-spot mistakes, depending on the technology used and how blindly we trust it.Benj EdwardsSenior AI ReporterBenj EdwardsSenior AI Reporter Benj Edwards is Ars Technica's Senior AI Reporter and founder of the site's dedicated AI beat in 2022. He's also a tech historian with almost two decades of experience. In his free time, he writes and records music, collects vintage computers, and enjoys nature. He lives in Raleigh, NC. 17 Comments

Lisa Morgan, Freelance WriterMarch 11, 20258 Min Read Cagkan Sayin via Alamy StockAccording to Pitchbook News, 35.7% of venture capital investments in 2024 were made in AI and ML startups, which isnt surprising given Big Techs investment in the same space. Meanwhile, EY reported that in Q4 2024, AI startups represented 60% of investments. While AI will continue to be a major focus area, under the Trump administration, theres a more bullish attitude about crypto, so some VCs are adding those companies to their portfolios while others are investing in space tech and ESG.If you increasingly think about agents interacting with one another without necessarily humans in the loop, you need to have a payment layer that is as scalable as millions of agents. [This is] the biggest promise in what's happening on the crypto front, says Pascal Unger, managing partner of pre-seed VC firm focal.He also says software is moving from, or has moved from, a system of engagement to a system of intelligence. That system of intelligence enables AI-powered automation that is accelerating the pace of business.One of the ways that weve added to our diligence is trying to understand the current AI tech stack that founders are leveraging, what all theyve tried and how much people lean into constantly getting better and trying experimenting. Understanding why [founders] chose a certain tech stack says a lot about peoples willingness to lean into these things, which will eventually, at least we believe, translate into faster speed.Related:Execution speed is critical, so software abstraction enables founders and their teams to get to market faster with solutions.You move from incredibly in the weeds into a manager and orchestrator of different tools, and you spend a lot more time on oversight and thinking and structuring things so thats changed, says firms Unger. I won't be surprised if we get a useful warm up period for new products where it needs to get to know us, basically, and it gets better over the first three to six months. That will become a normal thing down the road. We [also] believe in the importance of nailing distribution. If you want to build a differentiator product from the start, you now need to nail the distribution even earlier.Pascal Unger, focalPascal Unger, focalThat approach results in faster early-stage growth, such as going from zero to more than $10 million in less than a year. Unger says that just three years ago, that sort of company performance would have been an outlier.Hemant Mohapatra, partner at Lightspeed Venture Partners, expects 2025 to be a lot like 2024, but the target segments will shift. As AI foundational layers stabilize and the winners become more apparent, the next phase will focus more on middleware and application startups.Related:At Lightspeed we have been very active globally across the AI stack -- foundation models, data, middleware, both horizontal and vertical apps, as well as AI enabled services, says Mohapatra. Our investment strategy remains the same: Find the most compelling founders with clear right to win in their categories and find and back them no matter which part of the world they come from.Mohapatra sees a lot of potential in AI and gaming in terms of creating immersive worlds dynamically, non-player characterswith advanced conversational capabilities and personalized gameplay.We are also very early in LLMs controlling various software tools to get complex jobs done and we will see a few mind-boggling demos here in 2025, says Mohapatra. I also expect consolidation across the middleware layer -- LLMs or scaled out AI companies looking to vertically integrate across data, tool orchestration or memory infrastructure.Daniel Kang, CEO and co-founder of Y-Combinator backed startup Flowbo and former VC at SoftBank Vision Fund expects the proportion of VC investments in AI to likely increase over time for a few reasonsRelated:Most of the AI discussion is not just about the technology, but its position as a platform. Foundational model companies like Anthropic and OpenAI have made it easy for anyone to use their technology to build on top of it, says Kang. Its akin to Apple creating a mobile platform powered through the iOS for others to build apps, shifting the platform from desktop to mobile. Thats why many wrapper companies will continue to emerge.Meanwhile, models are improving by the day, which is creating more opportunities for startups. However, the complexities will likely require greater precision, which is probably why many middle layers already exist between the foundational models and applications.On top of that, intense competition among model providers like OpenAI and Anthropic is creating a downward pricing pressure to provide compute at cost, says Kang. APIs costs have already been revised several times to be cheaper, while their app products, ChatGPT and Claude, have remained the same.He expects general purpose application as wrappers to lose their edge as models improve and companies require more specialized solutions for specific tasks or functions. That probably will mean verticalization and the middle layers adapting general models for specific requirements.For apps specifically, the primary differentiation will likely be around distribution and brand more than technology, as the models improve and costs fall, says Kang. Foundational models and middle layers will probably continue to differentiate through tech. While the timing is unclear, the rise of middle layers seems imminent.Where Else Money Is FlowingViktor Shpakovsky, general partner at the Beyond Earth Technologies VC firm thinks space tech is a smarter bet than AI or crypto.AI and crypto have dominated headlines, but both sectors are showing signs of overinflation and speculative hype. Meanwhile, space tech is emerging as the most promising industrial growth sector, driven by government backing, geopolitical competition and technological breakthroughs, says Shpakovsky. With Trump [in] office, defense and space budgets are set to increase dramatically. Elon Musk and SpaceX continue to push commercial space forward at an unprecedented pace. At the same time, the US-China space race is accelerating, forcing the US government to invest aggressively in private space companies. These factors make 2025 the defining moment to bet on space tech over speculative software trends.He further reasons that AI and crypto are overcrowded and over-valued. While AI has become VCs latest gold rush, he says inflated values and copycat startups are challenges. Moreover, every startup claims that they have AI, but few have defensible technology or clear revenue models.As for crypto, he says the boom-and-bust cycle is predictable, because the crypto industry follows a well-known pattern: hype-driven speculation, price surges, regulatory crackdowns, and mass failures.Both AI and crypto sectors are flooded with startups, leading to undifferentiated competition and thinning margins, says Shpakovsky. Meanwhile, space tech remains an underinvested frontier with clear industrial demand. Unlike AI and crypto, space tech is a government-backed industrial growth sector. This isnt just about launching rockets -- its about building trillion-dollar infrastructure for the next era of human civilization.Viktor Shpakovsky, Beyond Earth TechnologiesViktor Shpakovsky, Beyond Earth TechnologiesInstead of chasing the next overhyped AI startup, Beyond Earth Technologies focuses on industries where AI is just a tool, not the entire business model. Its portfolio is built around lunar infrastructure, space robotics, next-gen energy and propulsion, in-space manufacturing, satellite intelligence, and space situational awareness.Benson Chang managing partner at Epipelagic Ventures expects a shift in 2025 -- not away from AI -- but toward infrastructure, cybersecurity and pragmatic applications with clear revenue models.Crypto may regain traction, particularly where blockchain solves real inefficiencies, says Chang. Weve refined our investment playbook to prioritize capital efficiency, durable moats and execution over hype. We need more than cutting-edge tech -- we must show defensibility, go-to-market traction and strategic innovation. The bar for funding is higher, and investors are backing leaner, more resilient teams.Anton Chashchin, founder and CEO at private fintech group N7 Capital, expects AI to remain a dominant investment theme attracting major inflow, but he warns that VCs should not overlook macroeconomic factors including persistent inflation and high interest rates.The global economic forecast is not as positive as we would like it to be. A predicted slowdown in global economic growth will make VCs more selective, prioritizing more sustainable startups with clear paths to profitability and not just AI use, says Chashchin.He also expects a greater focus on crypto because the Trump Administration strongly emphasizes it, which has fueled market growth and optimism.Considering that the total crypto market cap is projected to exceed $4 trillion, digital assets are no longer a speculative thing but an asset class attracting capital, says Chashchin. The institutionalization of the sector is also accelerating, meaning VCs should not ignore this opportunity.He also sees greater investments in renewables and ESG.As alternative types of energy become cheaper and more competitive, more companies operating in this field will receive funding from VCs, says Chashchin. The ongoing surge in ESG investment, projected to reach $50 trillion in assets under management this year, also highlights the growing demand for climate-focused solutions. With governments worldwide prioritizing clean energy, venture capital is shifting towards startups that are developing clean energy technologies or building infrastructure for their construction.Bottom LineAI investments are expected to remain high in 2025, with greater emphasis on the middle and application layers. Meanwhile, VC firms are making other investments, such as in crypto, space tech, renewable energy, and ESG.Read more about:Technology StartupsAbout the AuthorLisa MorganFreelance WriterLisa Morgan is a freelance writer who covers business and IT strategy and emergingtechnology for InformationWeek. She has contributed articles, reports, and other types of content to many technology, business, and mainstream publications and sites including tech pubs, The Washington Post and The Economist Intelligence Unit. Frequent areas of coverage include AI, analytics, cloud, cybersecurity, mobility, software development, and emerging cultural issues affecting the C-suite.See more from Lisa MorganWebinarsMore WebinarsReportsMore ReportsNever Miss a Beat: Get a snapshot of the issues affecting the IT industry straight to your inbox.SIGN-UPYou May Also Like

As I was writing this article, it was perhaps not so coincidental that I took a break and made a phone call to a home appliance customer support line for a microwave that we owned.I soon found myself trapped in an automated agentic AI phone system with no way out and no way to reach a human agent. I finally gave up, and called a local appliance company, where a human salesman gave me the answer that I needed.The experience is common. There are millions of consumers who experience frustration with automated phone systems and chat services that have no way of routing them to the person (or function) that can help them resolve their issues.Companies know this, but its not stopping them from adopting agentic AI at breakneck speeds, as evidenced by a projected market growth for agentic AI of 43.8% CAGR (compound annual growth rate) between now and 2034. Its all the more reason for CIOs to get involved early with agentic AI to make sure that it works for people as well as for systems.Just What Is Agentic AI and How Does it Work?In a 2024 interview with the Harvard Business Review, Enver Cetin, an AI expert at global experience engineering firm Ciklum, said, [Agentic AI] refers to AI systems and models that can act autonomously to achieve goals without the need for constant human guidance. The agentic AI system understands what the goal or vision of the user is and the context to the problem they are trying to solve.Related:Agentic AI uses a combination of machine learning (ML), natural language processing (NLP) and automation to do this. Its mission to make decisions and act on them.Companies can design agentic AI systems that require a final human authorization for some decisions, or they can make the agentic AI completely autonomous, so it makes decisions on its own.Most agentic AI adoptions are being sponsored and funded by end-user departments, which suggests that IT may or may not be in on the initial evaluations and buy decisions.Gartner cites an early example of how agentic AI can be deployed in retail.AI-enabled machine customers -- or nonhuman economic actors that obtain goods and services in exchange for payment --are examples of increasingly common intelligent agents. In the near future, they will make optimized decisions on behalf of human customers based on preset rules and will quickly evolve toward greater autonomy and inferring of needs.So, knowing that agentic AI is coming, and that IT might also be the last to know about an agentic AI buy decision, what should CIOs be doing?Related:Key CIO Points for Agentic AIWork up ITs agentic AI strategy now. Agentic AI has enormous potential. It can automate rote business operations and decision making, and IT needs to strategize for it.In a sense, agentic AI and what it can do has already been known in previous incarnations, such as automated loan decisioning software that has existed and functioned capably in bank lending departments for decades. However, now the needle is moving toward more autonomy. Business users will decide where they want to use agentic AI, but it will be ITs responsibility to ask the questions about system and process integration, governance and security that will enable agentic AI to be used safely and to best advantage.In this environment, an immediate CIO goal should be to participate with users in agentic AI strategy discussions so that best use business cases can be identified. Then, there should be a collaborative strategy with users and IT that takes into account not only business process streamlining and automation, but process exception handling, process and system integration. Plus, they should address user and IT training, security and governance. Although agentic AI will be driven by users, this is no time for IT to take a back seat.Related:Discuss security and failover. Your sales department might fund and adopt agentic AI to autonomously execute the mechanisms of product ordering, but what happens if a bad actor penetrates the agentic AI and locks it down for ransom or, what if that bad actor invades agentic AI software and injects malware or faulty algorithms that compromise and endanger the function?The sales team will quickly pivot to IT to fix these issues, so CIOs should be proactively querying sales and agentic AI vendors about the types of security that come with the agentic AI. There should be a company review of the AI to ensure that it complies with corporate security and governance standards. Questions should also be asked as to whether there is a failover to a human agent if the agentic AI fails or sputters. There should be a defined failover procedure in the company disaster recovery plan that provides for human ability to override or take over from agentic AI if that becomes necessary.Be prepared for project inclusion, whether you want it or not! IT may not be involved in initial agentic AI purchase decisions, but it will surely be pulled into agentic AI projects, because the AI wont get very far if it isnt integrated with other corporate systems.Accordingly, IT should ensure that user-IT agentic AI project discussions focus on system integration, and on the clear definition of a project test bed for agentic AI integration into business processes themselves.A successful business process integration addresses user training and readiness for a new technology, and what will happen if the agentic AI fails or begins to make poor decisions.CIOs shouldnt shy away from insisting that these process-oriented elements are tasked in agentic AI projects, because if anything goes wrong after the technology is placed into production, it will likely be blamed on the system.

This is today's edition ofThe Download,our weekday newsletter that provides a daily dose of what's going on in the world of technology. Two new measures show where AI models fail on fairness Whats new: A new pair of AI benchmarks could help developers reduce bias in AI models, potentially making them fairer and less likely to cause harm. The benchmarks evaluate AI systems based on their awareness of different scenarios and contexts. They could offer a more nuanced way to measure AIs bias and its understanding of the world.Why it matters: The researchers were inspired to look into the problem of bias after witnessing clumsy missteps in previous approaches, demonstrating how ignoring differences between groups may in fact make AI systems less fair. But while these new benchmarks could help teams better judge fairness in AI models, actually fixing them may require some other techniques altogether. Read the full story. Scott J Mulligan AGI is suddenly a dinner table topic The concept of artificial general intelligencean ultra-powerful AI system we dont have yetcan be thought of as a balloon, repeatedly inflated with hype during peaks of optimism (or fear) about its potential impact and then deflated as reality fails to meet expectations. Over the past week, lots of news went into inflating that AGI balloon, including the launch of a new, seemingly super-capable AI agent called Manus, created by a Chinese startup. Read our story to learn whats happened, and why it matters.James ODonnell This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here. The must-reads Ive combed the internet to find you todays most fun/important/scary/fascinating stories about technology. 1 The US has rebranded its immigration app with a self-deport functionIts a bid to encourage people living illegally to leave the country voluntarily. (AP News) + If they fail to self-report, undocumented migrants could face harsher consequences. (BBC)+ But immigrants should think very carefully before trusting the app. (The Guardian)+ The app was previously used to schedule asylum appointments. (MIT Technology Review)2 DOGE is scrabbling around for some wins The growing backlash against its clumsy cuts puts DOGEs top brass under pressure. (WP $)+ Biomedical research cuts would affect both elite and less-wealthy universities. (Undark)+ The agency is causing chaos within social securitys offices. (New Yorker $)+ The next phase? Handing over decisions to machines. (The Atlantic $)3 Donald Trump isnt a fan of the CHIPS Act Even though the law is designed to support chip manufacturing in the US. (NYT $)+ Heres what is at stake if he follows through on his threats to scrap it. (Bloomberg $)4 Elon Musk claims a cyber attack on X came from the Ukraine area But the billionaire, who is a fierce critic of Ukraine, hasnt provided any evidence. (FT $)+ The platform buckled temporarily under the unusually powerful attack. (Reuters)+ Cyber experts arent convinced, however. (AP News)5 AI-powered PlayStation characters are on the horizon Sony is testing out AI avatars that can hold conversations with players. (The Verge)+ How generative AI could reinvent what it means to play. (MIT Technology Review)6 DeepSeeks founder isnt fussed about making a quick buck Liang Wenfeng is turning down big investment offers in favor of retaining the freedom to make his own decisions. (WSJ $)+ Chinas tech optimism is at an all-time high. (Bloomberg $)+ How DeepSeek ripped up the AI playbookand why everyones going to follow its lead. (MIT Technology Review) 7 The rain is full of pollutants, including microplasticsAnd you thought acid rain was bad. (Vox) 8 An all-electric seaglider is being tested in Rhode IslandIt can switch seamlessly between floating and flying. (New Scientist $) + These aircraft could change how we fly. (MIT Technology Review)9 Tesla Cybertruck owners have formed an emotional support group One member is pushing for Cybertruck abuse to be treated as hate crimes. (Fast Company $)10 Theres only one good X account left Step forward Joyce Carol Oates. (The Guardian)Quote of the day There is no more asylum. US immigration officials tell a businessman seeking legitimate asylum that he cant enter the country just days after Donald Trump took office, the Washington Post reports. The big storyNext slide, please: A brief history of the corporate presentation August 2023 PowerPoint is everywhere. Its used in religious sermons; by schoolchildren preparing book reports; at funerals and weddings. In 2010, Microsoft announced that PowerPoint was installed on more than a billion computers worldwide. But before PowerPoint, 35-millimeter film slides were king. They were the only medium for the kinds of high-impact presentations given by CEOs and top brass at annual meetings for stockholders, employees, and salespeople. Known in the business as multi-image shows, these presentations required a small army of producers, photographers, and live production staff to pull off. Read this story to delve into the fascinating, flashy history of corporate presentations.Claire L. EvansWe can still have nice things A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or skeet 'em at me.)+ Heres how to prevent yourself getting a crick in the neck during your next flight.+ I would love to go on all of these dreamy train journeys.+ This Singaporean chocolate cake is delightfully simple to make.+ Meet Jo Nemeth, the woman who lives entirely without money.

Bird flu has caused egg prices to surge and has impacted the availability of eggs.Substitutes, such as applesauce, can be used in baking instead.I made brownies using applesauce, tofu, chia seeds, mayonnaise, and aquafaba to varying results.With the price of eggs reaching historic highs and supply wavering due to the bird flu, it could be worth considering egg substitutes when making meals at home.I tried making a classic box of Betty Crocker brownies with different substitutes to see which is best for baking.I used five common egg replacements recommended by vegans and home cooks: applesauce, silken tofu, mayonnaise, chia seeds, and aquafaba, which is the liquid found in a can of beans or legumes, such as chickpeas.Through this experiment, I hoped to find a cheaper alternative to eggs that could still hold up in the recipe.Here's how the five egg alternatives compared.To easily compare the results of the different egg substitutes, I used the same brownie mix for each batch: the Betty Crocker's Fudge mix.The classic mix offers a fudgy and rich brownie. Kristine Villarroel/Business Insider For this test, I used Betty Crocker's Fudge brownie mix. Each box was $2.99.The classic fudge mix calls for 1/2 a cup of oil, 2 tablespoons of water, and two eggs.The first egg alternative I tried was applesauce.Applesauce is a common replacement for eggs in baked goods. Kristine Villarroel/Business Insider One 24-ounce jar of Mott's applesauce was $2.Since applesauce works as a binder and an emulsifier, according to Business Insider's egg replacement chart, it should provide moisture and body to baked goods like brownies.I had used applesauce in baking, but never with boxed mixes.Applesauce works well holding ingredients together, but doesn't help with helping baked goods rise. Kristine Villarroel/Business Insider I had previously made oat cookies with applesauce, but I had never tried it in a box mix.To replace the two eggs in the brownies, I used 100 grams of applesauce, which is just over a 1/3 cup.The mix appeared runny and lumpy.The applesauce texture showed in the brownie mix. Kristine Villarroel/Business Insider I first mixed the applesauce with water and then added the oil, but it didn't seem to fully incorporate.When I poured the batter into a lined baking pan, the mix still had some oil floating on top.In the pan, I could see some of the oil creep up in the corners. Kristine Villarroel/Business Insider Since it wasn't fully emulsified, the applesauce mix seemed to separate a bit in the pan. After pouring it in, I saw some of the oil appearing in the corners.After baking, the same problem persisted I could see the oil bubbling up from the brownie.The oil seemed to separate and float at the top of the brownie. Kristine Villarroel/Business Insider I cooked each batch of brownies in this test at 350 degrees Fahrenheit for 35 minutes or more as needed.At the 35-minute mark, the applesauce brownies still looked undercooked, so I left them in the oven for an additional 10 minutes. Even then, the brownies still seemed a little undercooked when I cut them.The applesauce brownies felt oily and flat.The brownies felt a little undercooked. Kristine Villarroel/Business Insider Despite being cooked for the appropriate time, the brownies seemed undercooked on the inside maybe they could've benefited from staying in the oven a few extra minutes so the mix could solidify a bit more.However, when it came to the taste, they were fudgy and rich and didn't have any apple taste.The second egg alternative I tried was silken tofu.Tofu is a protein-rich replacement. Kristine Villarroel/Business Insider One 16-ounce container of tofu was $2.99.I used 160 grams to replace two eggs just under a third of the pack which added about 7 grams of protein to the brownies.The tofu intrigued me the most.I had never used tofu in baking. Kristine Villarroel/Business Insider While I'm familiar with tofu from being a vegan in the past, I had never used it in baking, so I was intrigued by its potential.The silken tofu was very soft, and it was a little challenging to drain the water from the container without the tofu also falling out.The tofu emulsified better than the applesauce.After mixing thoroughly, the tofu blended nicely in the mix. Kristine Villarroel/Business Insider I first mixed the tofu with the 2 tablespoons of water and then added the oil. The tofu seemed to emulsify a lot better than the applesauce.At first, white clots of tofu were a little intimidating, but after mixing it more, they blended nicely.The brownie batter was a little thicker than when I had used applesauce.The tofu made for a thicker, smoother batter. Kristine Villarroel/Business Insider The mix was thicker than the applesauce, but it was also smoother.The tofu brownie looked promising.The brownie looked shinier and more cooked. Kristine Villarroel/Business Insider Out of the oven, the tofu brownie had a nice shiny layer on top.It had also risen a lot more than the applesauce brownies and didn't look as oily.The tofu brownies were light, although a bit crumbly.The tofu had an almost cake-like texture. Kristine Villarroel/Business Insider Having risen a little more than the applesauce brownies, the tofu brownies had more of a cake-like texture and were a lot lighter.They were a little crumbly not dry but I didn't mind it too much because of the nice lightness the brownie had.The third egg alternative I tried was mayonnaise.Mayonnaise is made from eggs and oil. Kristine Villarroel/Business Insider Mayonnaise, which is made from eggs and oil, made sense as an egg replacement, although I wasn't very fond of the idea.One 8-ounce jar of Hellmann's mayonnaise was $3.99, making this alternative the most expensive per serving.I wasn't too excited to use mayonnaise in baking.I was the most skeptical about the mayonnaise. Kristine Villarroel/Business Insider While I can enjoy mayonnaise in savory dishes, the idea of using it for brownies was a little daunting. I wasn't looking forward to it, but I was open to having my mind changed.I used 6 tablespoons of mayo to replace the two eggs.The mix was the thickest.The mayonnaise made the brownie batter thicker than the rest of the alternatives. Kristine Villarroel/Business Insider I first mixed the mayonnaise with the water and then added the oil, and it seemed to emulsify really well.The mayonnaise brownies didn't look promising.The mix seemed to separate. Kristine Villarroel/Business Insider Pouring it in the pan, the mix did not look as smooth as when I had used other egg alternatives. It was also the thickest, so I had to spread it out.The brownies had trouble baking.The brownies took twice as long to cook. Kristine Villarroel/Business Insider When I first checked on the brownies at the 35-minute mark, the mix was still completely liquid.I let them cook until they solidified for an extra 30 minutes, 10 minutes at a time, and checked until they looked done.When I pulled them out of the oven, oil was bubbling through the flat top.The mayonnaise brownie was my least favorite.I thought it had a weird flavor, but it could've been more of a mental block. Kristine Villarroel/Business Insider The mayonnaise brownie was extremely oily and a little too chewy and hard.I also thought it had a bit of a weird flavor, but that might have been more of a mental block on my part.Next up: chia seeds.A two-pound bag of the seeds was $8.88. Kristine Villarroel/Business Insider The chia seeds seemed expensive at $8.88 for a 2-pound bag, which was the only one I could find at the store.Still, the large bag meant the cost per serving was actually lower than the mayonnaise.I was familiar with using them in baking.I had baked with chia seeds before and usually found it to make recipes more crumbly. Kristine Villarroel/Business Insider I had used chia seeds in baking before, so I knew going in that it usually meant recipes turned out more crumbly than expected.I first mixed 2 tablespoons of seeds with 6 tablespoons of water to make a "chia egg" it took about five minutes for the mixture to settle in a gelatinous texture.I then mixed it with water and then the oil.The mix was a little thick but smooth.You could still see the seeds in the mix. Kristine Villarroel/Business Insider Although you could still see the seeds, the mix itself was smooth and not too dry.The mix was runny enough to spread out on the pan.Unlike the mayonnaise batter, the chia seeds still made for a smooth mix. Kristine Villarroel/Business Insider Although textured, the mix was still smooth and runny enough to cover the pan without me spreading it.The chia brownies looked lighter and crispier than some of the other batches.The chia brownies had a nice crispy edge. Kristine Villarroel/Business Insider When I took them out of the oven, the brownies didn't seem to have any oil bubbling through and had a nice crispy edge and top.The chia brownies were crunchy and fudgy.The chia seeds added a crunch to the brownies. Kristine Villarroel/Business Insider With a nice crispy top and fudgy bottom, the chia brownies were my favorites by far. They were the perfect balance of rich and crunchy.The chia seeds also added a fun crunch to the brownie, and I really enjoyed the texture.The fifth and final egg alternative I tried was aquafaba.I had heard of aquafaba as an egg white replacement but had never used it. Kristine Villarroel/Business Insider Aquafaba, the liquid leftover from cooking chickpeas, is supposed to help with emulsifying, binding, and leavening, making it the most complete replacement, at least on paper.One can of chickpeas was $1.25, making this the cheapest alternative. I used 6 tablespoons of the liquid to replace two eggs.I was excited to try aquafaba, although a bit scared.The chickpea liquid was thick and runny, like watered-down egg whites. Kristine Villarroel/Business Insider I had never used aquafaba, and while I had heard a lot of good things about it, I was a bit uncertain about how these would turn out.Filtering out the liquid, I had to remove some chickpea skins from the aquafaba, which had a thick, watery texture, like watered-down egg whites.When mixed, the liquid turned a white color.It also became lighter and egg-white-like. Kristine Villarroel/Business Insider I first mixed the aquafaba with water and then added the oil. While mixing it, the liquid became white and lighter. I was afraid of over-beating it, so I stopped once it felt like a liquid egg mixture.This mixture was the runniest of them all.The aquafaba mixture was thin and runny. Kristine Villarroel/Business Insider While it emulsified well and the oil didn't seem to separate, the mixture was very runny, more than the other alternatives.