A marketing campaign brief obtained by WIRED reveals how the Chinese app is trying to capitalize on its sudden international popularity.

Ask people building generative AI what generative AI is good for right nowwhat theyre really fired up aboutand many will tell you: coding.Thats something thats been very exciting for developers, Jared Kaplan, chief scientist at Anthropic, told MIT Technology Review this month: Its really understanding whats wrong with code, debugging it.Copilot, a tool built on top of OpenAIs large language models and launched by Microsoft-backed GitHub in 2022, is now used by millions of developers around the world. Millions more turn to general-purpose chatbots like Anthropics Claude, OpenAIs ChatGPT, and Google DeepMinds Gemini for everyday help.Today, more than a quarter of all new code at Google is generated by AI, then reviewed and accepted by engineers, Alphabet CEO Sundar Pichai claimed on an earnings call in October: This helps our engineers do more and move faster. Expect other tech companies to catch up, if they havent already.Its not just the big beasts rolling out AI coding tools. A bunch of new startups have entered this buzzy market too. Newcomers such as Zencoder, Merly, Cosine, Tessl (valued at $750 million within months of being set up), and Poolside (valued at $3 billion before it even released a product) are all jostling for their slice of the pie. It actually looks like developers are willing to pay for copilots, says Nathan Benaich, an analyst at investment firm Air Street Capital: And so code is one of the easiest ways to monetize AI.Such companies promise to take generative coding assistants to the next level. Instead of providing developers with a kind of supercharged autocomplete, like most existing tools, this next generation can prototype, test, and debug code for you. The upshot is that developers could essentially turn into managers, who may spend more time reviewing and correcting code written by a model than writing it from scratch themselves.But theres more. Many of the people building generative coding assistants think that they could be a fast track to artificial general intelligence (AGI), the hypothetical superhuman technology that a number of top firms claim to have in their sights.The first time we will see a massively economically valuable activity to have reached human-level capabilities will be in software development, says Eiso Kant, CEO and cofounder of Poolside. (OpenAI has already boasted that its latest o3 model beat the companys own chief scientist in a competitive coding challenge.)Welcome to the second wave of AI coding.Correct codeSoftware engineers talk about two types of correctness. Theres the sense in which a programs syntax (its grammar) is correctmeaning all the words, numbers, and mathematical operators are in the right place. This matters a lot more than grammatical correctness in natural language. Get one tiny thing wrong in thousands of lines of code and none of it will run.The first generation of coding assistants are now pretty good at producing code thats correct in this sense. Trained on billions of pieces of code, they have assimilated the surface-level structures of many types of programs.But theres also the sense in which a programs function is correct: Sure, it runs, but does it actually do what you wanted it to? Its that second level of correctness that the new wave of generative coding assistants are aiming forand this is what will really change the way software is made.Large language models can write code that compiles, but they may not always write the program that you wanted, says Alistair Pullen, a cofounder of Cosine. To do that, you need to re-create the thought processes that a human coder would have gone through to get that end result.The problem is that the data most coding assistants have been trained onthe billions of pieces of code taken from online repositoriesdoesnt capture those thought processes. It represents a finished product, not what went into making it. Theres a lot of code out there, says Kant. But that data doesnt represent software development.What Pullen, Kant, and others are finding is that to build a model that does a lot more than autocompleteone that can come up with useful programs, test them, and fix bugsyou need to show it a lot more than just code. You need to show it how that code was put together.In short, companies like Cosine and Poolside are building models that dont just mimic what good code looks likewhether it works well or notbut mimic the process that produces such code in the first place. Get it right and the models will come up with far better code and far better bug fixes.BreadcrumbsBut you first need a data set that captures that processthe steps that a human developer might take when writing code. Think of these steps as a breadcrumb trail that a machine could follow to produce a similar piece of code itself.Part of that is working out what materials to draw from: Which sections of the existing codebase are needed for a given programming task? Context is critical, says Zencoder founder Andrew Filev. The first generation of tools did a very poor job on the context, they would basically just look at your open tabs. But your repo [code repository] might have 5000 files and theyd miss most of it.Zencoder has hired a bunch of search engine veterans to help it build a tool that can analyze large codebases and figure out what is and isnt relevant. This detailed context reduces hallucinations and improves the quality of code that large language models can produce, says Filev: We call it repo grokking.Cosine also thinks context is key. But it draws on that context to create a new kind of data set. The company has asked dozens of coders to record what they were doing as they worked through hundreds of different programming tasks. We asked them to write down everything, says Pullen: Why did you open that file? Why did you scroll halfway through? Why did you close it? They also asked coders to annotate finished pieces of code, marking up sections that would have required knowledge of other pieces of code or specific documentation to write.Cosine then takes all that information and generates a large synthetic data set that maps the typical steps coders take, and the sources of information they draw on, to finished pieces of code. They use this data set to train a model to figure out what breadcrumb trail it might need to follow to produce a particular program, and then how to follow it.Poolside, based in San Francisco, is also creating a synthetic data set that captures the process of coding, but it leans more on a technique called RLCEreinforcement learning from code execution. (Cosine uses this too, but to a lesser degree.)RLCE is analogous to the technique used to make chatbots like ChatGPT slick conversationalists, known as RLHFreinforcement learning from human feedback. With RLHF, a model is trained to produce text thats more like the kind human testers say they favor. With RLCE, a model is trained to produce code thats more like the kind that does what it is supposed to do when it is run (or executed).Gaming the systemCosine and Poolside both say they are inspired by the approach DeepMind took with its game-playing model AlphaZero. AlphaZero was given the steps it could takethe moves in a gameand then left to play against itself over and over again, figuring out via trial and error what sequence of moves were winning moves and which were not.They let it explore moves at every possible turn, simulate as many games as you can throw compute atthat led all the way to beating Lee Sedol, says Pengming Wang, a founding scientist at Poolside, referring to the Korean Go grandmaster that AlphaZero beat in 2016. Before Poolside, Wang worked at Google DeepMind on applications of AlphaZero beyond board games, including FunSearch, a version trained to solve advanced math problems.When that AlphaZero approach is applied to coding, the steps involved in producing a piece of codethe breadcrumbsbecome the available moves in a game, and a correct program becomes winning that game. Left to play by itself, a model can improve far faster than a human could. A human coder tries and fails one failure at a time, says Kant. Models can try things 100 times at once.A key difference between Cosine and Poolside is that Cosine is using a custom version of GPT-4o provided by OpenAI, which makes it possible to train on a larger data set than the base model can cope with, but Poolside is building its own large language model from scratch.Poolsides Kant thinks that training a model on code from the start will give better results than adapting an existing model that has sucked up not only billions of pieces of code but most of the internet. Im perfectly fine with our model forgetting about butterfly anatomy, he says.Cosine claims that its generative coding assistant, called Genie, tops the leaderboard on SWE-Bench, a standard set of tests for coding models. Poolside is still building its model but claims that what it has so far already matches the performance of GitHubs Copilot.I personally have a very strong belief that large language models will get us all the way to being as capable as a software developer, says Kant.Not everyone takes that view, however.Illogical LLMsTo Justin Gottschlich, the CEO and founder of Merly, large language models are the wrong tool for the jobperiod. He invokes his dog: No amount of training for my dog will ever get him to be able to code, it just wont happen, he says. He can do all kinds of other things, but hes just incapable of that deep level of cognition.Having worked on code generation for more than a decade, Gottschlich has a similar sticking point with large language models. Programming requires the ability to work through logical puzzles with unwavering precision. No matter how well large language models may learn to mimic what human programmers do, at their core they are still essentially statistical slot machines, he says: I cant train an illogical system to become logical.Instead of training a large language model to generate code by feeding it lots of examples, Merly does not show its system human-written code at all. Thats because to really build a model that can generate code, Gottschlich argues, you need to work at the level of the underlying logic that code represents, not the code itself. Merlys system is therefore trained on an intermediate representationsomething like the machine-readable notation that most programming languages get translated into before they are run.Gottschlich wont say exactly what this looks like or how the process works. But he throws out an analogy: Theres this idea in mathematics that the only numbers that have to exist are prime numbers, because you can calculate all other numbers using just the primes. Take that concept and apply it to code, he says.Not only does this approach get straight to the logic of programming; its also fast, because millions of lines of code are reduced to a few thousand lines of intermediate language before the system analyzes them.Shifting mindsetsWhat you think of these rival approaches may depend on what you want generative coding assistants to be.In November, Cosine banned its engineers from using tools other than its own products. It is now seeing the impact of Genie on its own engineers, who often find themselves watching the tool as it comes up with code for them. You now give the model the outcome you would like, and it goes ahead and worries about the implementation for you, says Yang Li, another Cosine cofounder.Pullen admits that it can be baffling, requiring a switch of mindset. We have engineers doing multiple tasks at once, flitting between windows, he says. While Genie is running code in one, they might be prompting it to do something else in another.These tools also make it possible to protype multiple versions of a system at once. Say youre developing software that needs a payment system built in. You can get a coding assistant to simultaneously try out several different optionsStripe, Mango, Checkoutinstead of having to code them by hand one at a time.Genie can be left to fix bugs around the clock. Most software teams use bug-reporting tools that let people upload descriptions of errors they have encountered. Genie can read these descriptions and come up with fixes. Then a human just needs to review them before updating the code base.No single human understands the trillions of lines of code in todays biggest software systems, says Li, and as more and more software gets written by other software, the amount of code will only get bigger.This will make coding assistants that maintain that code for us essential. The bottleneck will become how fast humans can review the machine-generated code, says Li.How do Cosines engineers feel about all this? According to Pullen, at least, just fine. If I give you a hard problem, youre still going to think about how you want to describe that problem to the model, he says. Instead of writing the code, you have to write it in natural language. But theres still a lot of thinking that goes into that, so youre not really taking the joy of engineering away. The itch is still scratched.Some may adapt faster than others. Cosine likes to invite potential hires to spend a few days coding with its team. A couple of months ago it asked one such candidate to build a widget that would let employees share cool bits of software they were working on to social media.The task wasnt straightforward, requiring working knowledge of multiple sections of Cosines millions of lines of code. But the candidate got it done in a matter of hours. This person who had never seen our code base turned up on Monday and by Tuesday afternoon hed shipped something, says Li. We thought it would take him all week. (They hired him.)But theres another angle too. Many companies will use this technology to cut down on the number of programmers they hire. Li thinks we will soon see tiers of software engineers. At one end there will be elite developers with million-dollar salaries who can diagnose problems when the AI goes wrong. At the other end, smaller teams of 10 to 20 people will do a job that once required hundreds of coders. It will be like how ATMs transformed banking, says Li.Anything you want to do will be determined by compute and not head count, he says. I think its generally accepted that the era of adding another few thousand engineers to your organization is over.Warp drives Indeed, for Gottschlich, machines that can code better than humans are going to be essential. For him, thats the only way we will build the vast, complex software systems that he thinks we will eventually need. Like many in Silicon Valley, he anticipates a future in which humans move to other planets. Thats only going to be possible if we get AI to build the software required, he says: Merlys real goal is to get us to Mars.Gottschlich prefers to talk about machine programming rather than coding assistants, because he thinks that term frames the problem the wrong way. I dont think that these systems should be assisting humansI think humans should be assisting them, he says. They can move at the speed of AI. Why restrict their potential?Theres this cartoon called The Flintstones where they have these cars, but they only move when the drivers use their feet, says Gottschlich. This is sort of how I feel most people are doing AI for software systems.But what Merlys building is, essentially, spaceships, he adds. Hes not joking. And I dont think spaceships should be powered by humans on a bicycle. Spaceships should be powered by a warp engine.If that sounds wildit is. But theres a serious point to be made about what the people building this technology think the end goal really is.Gottschlich is not an outlier with his galaxy-brained take. Despite their focus on products that developers will want to use today, most of these companies have their sights on a far bigger payoff. Visit Cosines website and the company introduces itself as a Human Reasoning Lab. It sees coding as just the first step toward a more general-purpose model that can mimic human problem-solving in a number of domains.Poolside has similar goals: The company states upfront that it is building AGI. Code is a way of formalizing reasoning, says Kant.Wang invokes agents. Imagine a system that can spin up its own software to do any task on the fly, he says. If you get to a point where your agent can really solve any computational task that you want through the means of softwarethat is a display of AGI, essentially.Down here on Earth, such systems may remain a pipe dream. And yet software engineering is changing faster than many at the cutting edge expected.Were not at a point where everythings just done by machines, but were definitely stepping away from the usual role of a software engineer, says Cosines Pullen. Were seeing the sparks of that new workflowwhat it means to be a software engineer going into the future.

A prolific and generally accurate leaker is predicting that the next iPad Air will use the M3 chip, countering rumors of a jump up to the M4 chip.The base of an iPad Air displayApple is anticipated to launch an update to the iPad Air in the spring, with a refresh upgrading its Apple Silicon performance. If a leaker is to be believed, that jump may not be as big as rumors have claimed.An image privately released on social media by Evan Blass, known as "evleaks," appears to be listings for the next iPad and iPad Air. While a lot of the information is blurred, the description of the iPad Air seems to refer to its next chip. Rumor Score: Possible Continue Reading on AppleInsider | Discuss on our Forums

In an interesting twist on the speedruning competition formula, the finalists for a recent Astro Bot speedruning tournament hosted by PlayStation had to compete against each other in an entirely new level that has never been showcased before.In order to be able to complete the level, both competitors had access to the slo-mo powerup, which saw heavy use throughout the finalists attempts at the level, which was seemingly designed around using slo-mo as well as the flower umbrella. The level the finalists had to compete in was themed around water and ice, with both players having to deal with obstacles like slippery ice floors and platforms that could easily shatter.Ultimately, Team Circle was able to win the tournament with a fastest time of 34.51.6 seconds, while Team Square narrowly lost with a time of 34.84.1 seconds. It is worth noting that both players had zero knowledge of the level used in the finals, and would likely had gotten even faster completion times if they had been allowed any sort of time to practice.As for its part, developer Team Asobi hasnt made any announcement of this level being released through an update to the game to let everyone have a shot at setting their own records.The last major update to hit Astro Bot was back in December, when the game got the Christmas-themed Winter Wonder level. The difficulty of the level wasnt set too high, as it was aimed towards both children as well as adults.Astro Bot came out exclusively on the PS5 last year. For more details about the game, check out our review. Astro Bot has also been incredibly successful, ending up being the best-reviewed title on the PS5 for 2024, and even managed to sell more than 1.5 million copies in nine weeks.

Polish mobile developer and publisher Huuuge Games is laying off 29 percent of its global workforce.The company confirmed the news in a note to investors last week and said it wants to "streamline operations" and focus on its core social casino business.The layoffs will impact 122 workers company-wide, including 89 employees based in Poland (representing 25 percent of staff in the region)."The workforce reduction is expected to be completed by the end of February 2025 and will not negatively impact the company's revenue," added Huuuge.Huuuge is the free-to-play specialist behind titles including Stars Slots, Billionaire Casino, and Dominos Party. The company explained the redundancies are designed to "improve agility" while reducing costs.With that goal in mind, Huuuge will also be shuttering its 'Huuuge Pods' (a cluster of small internal teams formed in 2023) across the Netherlands and Finland and discontinuing new game development.Huuuge said it remains profitable with $125 million in cash reserves as of Q3 2024. The company will continue to explore growth opportunities through potential M&A investments that "leverage its expertise in monetization and games-as-a-service."Company CEO Wojciech Wronowski said the decision to cut jobs is "never easy" but is indicative of Huuuge's commitment to "accelerating our ability to innovate."Huuuge remains a profitable company, and this marks a pivotal moment as we chart a path forward to deliver even greater value to our players, partners, and shareholders," he added.

Canons new app lets you live stream from multiple cameras, just not CanonsCanons new app lets you live stream from multiple cameras, just not Canons / The Live Switcher Mobile app turns iPhones and iPads into a mobile wireless studio that can stream from three devices.By Andrew Liszewski, a senior reporter who's been covering and reviewing the latest gadgets and tech since 2011, but has loved all things electronic since he was a kid. Jan 20, 2025, 3:05 PM UTCShare this storyCanons Live Switcher Mobile app is free for iOS and iPadOS devices. Screenshot: YouTubeCanon has announced a new live streaming app that can wirelessly connect to, and switch between, video feeds from three mobile devices. Its currently limited to iPhones and iPads running at least iOS and iPad0S 16. Android isnt supported, but an even stranger omission is that Live Switcher Mobile isnt compatible with any of Canons digital cameras that functionality will be added in the future, according to an FAQ.The app is available for download now and can be used for free, but with some limitations. Video quality maxes out at 720p, the output is watermarked, and the apps interface will occasionally be cluttered with ads but those arent presented to viewers. For $17.99 per month (which is over $215 per year) output quality gets a bump to 1080p, while ads and the watermark are removed. Theres no option to increase the video resolution past HD.The mobile device serving as a live streams primary camera also provides the switching interface. Image: CanonOne device running the app is designated as the host which provides the switching interface and serves as the primary camera. Up to two additional devices can be connected providing video feeds from alternate angles, or a live screen capture of a video game being played.Switching between feeds can be done manually, but if youd like to avoid poking a device thats serving as one of your cameras, the app can also be set to automatically switch between sources at set intervals of eight to 20 seconds. Multiple videos can be streamed simultaneously using a picture-in-picture effect, while text and image overlays can be added to personalize a livestream, add branding, or provide captioning.Canon says the app supports live streaming to YouTube, Facebook, and other platforms that support RTMP, but its comment display function is limited to those first two platforms.Although its functionality isnt as robust as the free OBS Studio app that also allows mobile devices to be used as video sources, Canons Live Switcher Mobile is potentially an easier alternative for aspiring influencers just getting started with live streaming.It may currently be limited to mobile devices, but the company has confirmed the app will support Canons digital camera connection in the future. With that added functionality, and maybe a bump to 4K streaming capabilities, its steep price tag could be easier to justify for those looking for a highly portable streaming solution.Most PopularMost Popular

The nearly open-source free Defold game engine just released Defold 1.9.7 and the highlight feature of this release is C# support! This isnt replacing Lua as the primary scripting language, in fact its not a scripting language at all, instead enabling you to easily make game extensions using the C# programming language with Defold. This language support joins other programming languages including Zig and C++. There was hesitation to add C# support to Defold, explained below:Q: I am concerned that adding C# support will have a negative impact on Defold. Should I be worried?Defold is NOT moving away from Lua as a the primary scripting language. C# support will be added as a new language for extensions. It will not impact the engine unless you choose to use C# extensions in your project.C# support will come at a price (executable size, runtime performance etc), but thats for the individual developer/studio to decide upon.As for C# itself, its a relatively minor change, since the extension system already support many languages (C/C++/Java/Objective-C/Zig). The SDKs will be kept in sync by generating the C# bindings. This will keep the bindings up-to-date with minimal effort.The Defold Foundation has previously been against adding C# support in Defold, but has changed opinion for a number of reasons:Studios and developer continue to request C# support.C# support has been scoped down to extensions only (i.e. low effort).The core engine will not be impacted.The C# APIs can be kept in in sync with minimal effort if they are generated.C# support will be based on DotNet 9 with NativeAOT, thus generating static libraries that the existing build pipeline can link against (just like any other Defold extension).Key LinksDefold 1.9.7 BetaDefold C# Support AnnouncementDefold GitHubDefold HomepageDefold Languages SampleYou can learn more about the Defold game engine, the 1.9.7 release and the new addition of C# support in the video below.

Swarm is an innovative open-source framework designed to explore the orchestration and coordination of multi-agent systems. It is developed and managed by the OpenAI Solutions team, and it provides a lightweight, ergonomic, and educational environment for developers to learn and experiment with agent-based systems. At its core, Swarm is built to facilitate the interaction of autonomous Agents, i.e., independent units capable of performing specific tasks, through streamlined handoffs and routine management. While primarily aimed at educational use, the framework introduces patterns and abstractions that make multi-agent orchestration more accessible and comprehensible. By focusing on simplicity and modularity, Swarm allows users to design workflows where Agents can collaborate, delegate tasks, and share contextual data seamlessly. OpenAIs Chat Completions API entirely powers it; Swarm operates statelessly, ensuring security and flexibility. With no official support or production readiness, Swarm is a learning platform.Core Components of SwarmSwarm is built on fundamental components that provide a strong foundation for flexibility and functionality. These components include:AgentsAgents are the primary units in Swarm, each representing an independent actor or step in a process. They include:Instructions: Define the Agents behavior or task.Functions: Specify actions the Agent can perform, including function calls.Handoffs: Allow the Agent to delegate its task to another Agent.Agents are initialized as follows:# pythonfrom swarm import Agentagent_a = Agent( name="Agent A", instructions="You are a general-purpose assistant.", functions=[] # Add any callable functions here)HandoffsHandoffs enable one Agent to pass control to another seamlessly. This allows specialized Agents to handle tasks better suited to their capabilities.# pythonagent_b = Agent( name="Agent B", instructions="You only provide answers in haikus.")agent_a = Agent( name="Agent A", instructions="Forward this task to Agent B.", functions=[lambda: agent_b] # Hand off to agent_b)Context VariablesContext variables store shared data across Agents, ensuring continuity in multi-agent workflows.# pythoncontext = {"user_name": "John"}response = client.run( agent=agent_a, messages=[{"role": "user", "content": "Who am I speaking with?"}], context_variables=context)How Swarm WorksAt its core, Swarm processes interactions using a structured loop implemented in its client.run() method. The loop involves the following steps:Message Processing: The current Agent processes the users message, which may generate a response or call a function.Function Execution: If the Agent includes function calls, these are executed, and the results are added to the conversation.Agent Switching: If the task requires another Agent, Swarm handles the handoff, ensuring seamless execution.Context Management: Context variables are updated throughout the interaction, ensuring shared data is accessible across Agents.Response Delivery: Swarm delivers the final response to the user after completing all steps.The basic workflow is illustrated below:# pythonfrom swarm import Swarm# Initialize the Swarm clientclient = Swarm()# Run the processresponse = client.run( agent=agent_a, messages=[{"role": "user", "content": "What can you do?"}])print(response.messages[-1]["content"])Usage of Swarm Code ImplementationInstallationSwarm can be installed directly from its GitHub repository:# bashpip install git+https://github.com/openai/swarm.gitBasic SetupSetting up Swarm involves importing the library, creating Agents, and running the interaction loop.# pythonfrom swarm import Swarm, Agent# Initialize Swarm clientclient = Swarm()# Define Agentsagent_a = Agent( name="Agent A", instructions="Provide general assistance.")agent_b = Agent( name="Agent B", instructions="Respond to all queries in poetic form.")# Interactionresponse = client.run( agent=agent_a, messages=[{"role": "user", "content": "Who am I speaking to?"}])print(response.messages[-1]["content"])Advanced FeaturesSwarm supports advanced features, including streaming responses and debugging.Streaming Responses:# pythonstream = client.run( agent=agent_a, messages=[{"role": "user", "content": "Stream a response"}], stream=True)for chunk in stream: print(chunk)Debugging:# pythonresponse = client.run( agent=agent_a, messages=[{"role": "user", "content": "Debug this process"}], debug=True)Download Colab NotebookConclusion:Swarm is an ergonomic, lightweight, and educational open-source framework that lets developers try out patterns and techniques essential for scalable agent orchestration. Although not meant for production, its focus on accessibility, modularity, and testability makes it a valuable resource for learning and prototyping. Its ability to support complex workflows through simple abstractions, such as Agents, handoffs, and context variables, allows developers to design effective solutions without being overwhelmed by technical complexities.Sourceshttps://github.com/openai/swarmhttps://colab.research.google.com/drive/1uFquKQvXLpKeP05OD507UFl8d0YvhM1t?authuser=1 Asif RazzaqAsif Razzaq is the CEO of Marktechpost Media Inc.. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. His most recent endeavor is the launch of an Artificial Intelligence Media Platform, Marktechpost, which stands out for its in-depth coverage of machine learning and deep learning news that is both technically sound and easily understandable by a wide audience. The platform boasts of over 2 million monthly views, illustrating its popularity among audiences. Meet 'Height':The only autonomous project management tool (Sponsored)