Cornerpond A game by foolsroomDev snapshot: Godot 4.5 dev 3By: Thaddeus Crews25 April 2025Pre-releaseIf the past snapshot was any indication, you might think that development updates aren’t slowing down anytime soon. And… You’d be right! Indeed, progress has been booming, and the amount of new features added even compared to the previous release is staggering; curating this selection was especially difficult. As always, new features mean new bugs in need of fixing, so we encourage everyone interested to get feedback and bug reports in as early as possible.Jump to the Downloads section, and give it a spin right now, or continue reading to learn more about improvements in this release. You can also try the Web editor or the Android editor for this release. If you are interested in the latter, please request to join our testing group to get access to pre-release builds.The cover illustration is from Cornerpond, a fishing game that takes place entirely in the corner of your desktop. It is developed by foolsroom. You can get the game on Steam and follow the developer on Twitter.HighlightsIn case you missed them, see the 4.5 dev 1 and 4.5 dev 2 release notes for an overview of some key features which were already in that snapshot, and are therefore still available for testing in dev 3.Screen reader supportAccessibility should be every developer’s top priority, full-stop. Someone being excluded from an experience for factors outside of their control is an area that video games and applications have the potential to circumvent entirely. It does, however, take a solid framework to allow such accommodations to be developed. To streamline this process for everyone—players and developers alike—our resident tech guru bruvzg took to the absolutely Herculean task of integrating AccessKit to Godot as a whole.GH-76829 was a project started two years ago, which progressed in bursts alongside the AccessKit framework. For the 4.5 release, we made it a priority to finalize, and thus we now merged this major feature with over 32,000 lines of code, after hundreds of comments with feedback/testing. Users are encouraged to look at the pull request for more information, as there’s no feasible way we could properly summarize these changes. Unsurprisingly, this was by far the change with the most ramifications of the entire snapshot, so much so that it’s already seen multiple fixes to address regressions (even one right when validating this snapshot), but it’s well worth it. After all, accessibility is our top priority!This is only the first but massive step towards making Godot more accessible. In particular for the editor, a lot more work will be needed to make it really usable, as well as integrate with accessibility frameworks for mobile or web platforms.Script backtracingIn any other snapshot, this would’ve been the main highlight, as adding backtracing to GDScript was among the most highly requested features from our users for years. Two brave souls, Mikael Hermansson (godot-jolt) and Juan Linietsky (up-and-coming developer), helmed this task and made this process possible with GH-91006. This will make it much easier for users to find the cause of warnings/errors that previously required manually hunting down bugs. Stack traces are now available in projects exported in release mode as well if the Debug > Settings > GDScript > Always Track Call Stacks project setting is enabled. This can make it easier for users to report issues in a way that developers can track down. Inspector section togglesAnother long-awaited feature, inspector section toggles, is now a part of the engine as of GH-105272. lodetrick has expanded editor functionality to what you see below: sections with their own dedicated checkbox to denote if they’re enabled.And more!There are too many exciting changes to list them all here, but here’s a curated selection:3D: Set correct position of node with Align Transform with View in orthographic view (GH-99099).Audio: Fix AudioStreamPlayer3D stereo panning issue (GH-104853).Buildsystem: Fix .sln project generation logic for Rider to support all OS and all C++ toolchains (GH-103405).Buildsystem: Update the Android NDK to the latest LTS version (r27c) (GH-105611).C#: Avoid unnecessary StringName allocations on not implemented virtual _Get and _Set method call (GH-104689).Core: Add create_id_for_path() to ResourceUID (GH-99543).Core: Add negative index to Array.remove_at and Array.insert (GH-83027).Core: Add thread safety to Object signals (GH-105453).Editor: Autocompletion: Don’t add parenthesis if Callable is expected (GH-96375).Editor: Fix exported Node/Resource variables resetting when extending script in the SceneTreeDock (GH-105148).Editor: Project manager: Add option to backup project when it will be changed (GH-104624).Editor: Support custom features in project settings dialog (GH-105307).Export: Use project settings overrides with the target preset features instead of current platform features (GH-71542).GDExtension: Optimize gdvirtual function layout (GH-104264).GUI: Add FoldableContainer (GH-102346).GUI: Add boolean toggle for middle-click to fire tab_close_pressed signal (GH-103024).GUI: Add separate minimize_disabled and maximize_disabled window flags (GH-105107).GUI: Add support for OEM Alt codes input (GH-93466).GUI: Implement SVGTexture auto-scalable with font oversampling (GH-105375).GUI: Make embed floating window respect Always On Top configuration (GH-103731).GUI: Replace global oversampling with overrideable per-viewport oversampling (GH-104872).Input: Add configuration option to disable Scroll Deadzone on Android (GH-96139).Input: Allow all tool modes to select (GH-87756).Plugin: Add maven publishing configuration for Godot tools (GH-104819).Porting: Android: Add new actions and enhancements to TouchActionsPanel (GH-105140).Porting: Android: Embed TouchActionsPanel directly into the editor UI (GH-105518).Rendering: Detect more pipeline settings at load time to avoid pipeline stutters (GH-105175).Rendering: Renderer: Reduce scope of mutex locks to prevent common deadlocks (GH-105138).XR: OpenXR: Request the XR_KHR_loader_init extension (GH-105445).Changelog115 contributors submitted 253 fixes for this release. See our interactive changelog for the complete list of changes since the previous 4.5-dev2 snapshot.This release is built from commit 28089c40c.DownloadsGodot is downloading...Godot exists thanks to donations from people like you. Help us continue our work:Make a DonationStandard build includes support for GDScript and GDExtension..NET build (marked as mono) includes support for C#, as well as GDScript and GDExtension.While engine maintainers try their best to ensure that each preview snapshot and release candidate is stable, this is by definition a pre-release piece of software. Be sure to make frequent backups, or use a version control system such as Git, to preserve your projects in case of corruption or data loss.Known issuesThere are currently no known issues introduced by this release.With every release, we accept that there are going to be various issues, which have already been reported but haven’t been fixed yet. See the GitHub issue tracker for a complete list of known bugs.Bug reportsAs a tester, we encourage you to open bug reports if you experience issues with this release. Please check the existing issues on GitHub first, using the search function with relevant keywords, to ensure that the bug you experience is not already known.In particular, any change that would cause a regression in your projects is very important to report (e.g. if something that worked fine in previous 4.x releases, but no longer works in this snapshot).SupportGodot is a non-profit, open source game engine developed by hundreds of contributors on their free time, as well as a handful of part and full-time developers hired thanks to generous donations from the Godot community. A big thank you to everyone who has contributed their time or their financial support to the project!If you’d like to support the project financially and help us secure our future hires, you can do so using the Godot Development Fund.Donate now

In the battle between two “agentic” coding tools — Anthropic’s Claude Code and OpenAI’s Codex CLI — the latter appears to be fostering more developer goodwill than the former. That’s at least partly because Anthropic has issued takedown notices to a developer trying to reverse-engineer Claude Code, which is under a more restrictive usage license than Codex CLI. Claude Code and Codex CLI are dueling tools that accomplish much of the same thing: allow developers to tap into the power of AI models running in the cloud to complete various coding tasks. Anthropic and OpenAI released them within months of each other — each company racing to capture valuable developer mindshare. The source code for Codex CLI is available under an Apache 2.0 license that allows for distribution and commercial use. That’s in contrast to Claude Code, which is tied to Anthropic’s commercial license. That limits how it can be modified without explicit permission from the company. Anthropic also “obfuscated” the source code for Claude Code. In other words, Claude Code’s source code isn’t readily available. When a developer de-obfuscated it and released the source code on GitHub, Anthropic filed a DMCA complaint — a copyright notification requesting the code’s removal. Developers on social media weren’t pleased by the move, which they said compared unfavorably with OpenAI’s rollout of Codex CLI. In the week or so since Codex CLI’s release, OpenAI has merged dozens of developer suggestions into the tool’s codebase, including one that lets Codex CLI tap AI models from rival providers — including Anthropic. Anthropic didn’t respond to a request for comment. To be fair to the lab, Claude Code is still in beta (and a bit buggy); it’s possible Anthropic will release the source code under a permissive license in the future. Companies have many reasons for obfuscating code, security considerations being one of them. It’s a somewhat surprising PR win for OpenAI, which in recent months has shied away from open source releases in favor of proprietary, locked-down products. It may be emblematic of a broader shift in the lab’s approach; OpenAI CEO Sam Altman earlier this year said he believed that the company has been on the “wrong side of history” when it comes to open source.

Hyena Edge’s success positions Liquid AI as one of the emerging players to watch in the evolving AI model landscape.Read More

Hyena Edge’s success positions Liquid AI as one of the emerging players to watch in the evolving AI model landscape.Read More

We knew Microsoft was about to launch Recall for real this time, and now the software maker is making it available to all Copilot Plus PCs. Recall, a feature that screenshots almost everything you do on a Copilot Plus PC, will be available today alongside an improved AI-powered Windows search interface and a new Click to Do feature that’s very similar to Google’s Circle to Search.Recall was originally supposed to launch at the same time as Copilot Plus PCs in June last year, but the feature was delayed following concerns raised by security researchers. Microsoft then planned to start publicly testing Recall in October, but pushed it back again to November to have more time to secure it further. Microsoft has now spent the past 10 months overhauling the security of Recall and making it an opt-in experience that you don’t have to enable if you’re concerned about the privacy implications.The Recall timeline lets you scroll through all the snapshots of your PC. Image: Microsoft“When we introduced Recall, we set out to address a common frustration: picking up where youleft off,“ explains Navjot Virk, corporate vice president of Windows Experiences at Microsoft. Recall is designed to improve how you search your PC, but taking snapshots that are categorized so it’s easy to search for vague memories instead of file names.I spent a few weeks testing Recall last year and found it was creepy, clever, and compelling. Technologically it’s a great improvement to the Windows search interface, because it can understand images and content in a much more natural way. But it does create a privacy minefield because you’re suddenly storing a lot more information on your PC usage, and you still need to manage blocked apps and websites carefully.Kevin Beaumont, one of the security researchers that first raised alarm bells over Recall, has been testing the final version recently and found that “Microsoft has made serious efforts to try to secure Recall.” The database is now encrypted, Recall attempts to filter sensitive data by default, and the feature is now an opt-in experience.Beaumont does note that filtering sensitive apps and websites can be hit-and-miss though, and occasionally even buggy. He also says that you can access Recall through a simple four-digit PIN unlock option with Windows Hello, instead of it forcing more secure facial recognition or a fingerprint. Microsoft’s Recall website claims “you must have at least one biometric sign-in option enabled for Windows Hello, either facial recognition or a fingerprint, to launch and use Recall.”The new AI-powered Windows search interface. Image: MicrosoftAlongside Recall, Windows search is also getting some AI improvements on Copilot Plus PCs today. You can now use the File Explorer, Windows search box, or settings with natural language queries. That means instead of searching for file names or specific settings, you can now describe images or documents and get results. If you’re looking for an image of a brown dog you know you have saved somewhere, you can just ask for “brown dog” rather than having to know the file name or date the image was created.Microsoft is also rolling out Click to Do today, which works a lot like Google’s Circle to Search. You activate it by using the Windows key + left mouse click, and it will provide actions for the text or images that are on your screen. This includes summarizing text or being able to quickly remove an object from an image.Click to Do lets you take actions on images and text. Image: MicrosoftRecall, the improved Windows search, and Click to Do will all be available today across all Copilot Plus PCs, but the text actions in Click to Do are currently limited to Qualcomm-powered devices, with AMD- and Intel-powered Copilot Plus PCs getting this feature “in the next few months.” Recall and Click to Do should be available in a variety of languages and regions, but Microsoft says both features won’t be available in EU countries and Iceland, Liechtenstein, and Norway until later this year.See More:

You might be wondering whether GenAI is just hype or external noise. I also thought this was hype, and I could sit this one out until the dust cleared. Oh, boy, was I wrong. GenAI has real-world applications. It also generates revenue for companies, so we expect companies to invest heavily in research. Every time a technology disrupts something, the process generally moves through the following phases: denial, anger, and acceptance. The same thing happened when computers were introduced. If we work in the software or hardware field, we might need to use GenAI at some point. In this article, I cover how to power your application with large Language Models (LLMs) and discuss the challenges I faced while setting up LLMs. Let’s get started. 1. Start by defining your use case clearly  Before jumping onto LLM, we should ask ourselves some questions a. What problem will my LLM solve? b. Can my application do without LLMc. Do I have enough resources and compute power to develop and deploy this application? Narrow down your use case and document it. In my case, I was working on a data platform as a service. We had tons of information on wikis, Slack, team channels, etc. We wanted a chatbot to read this information and answer questions on our behalf. The chatbot would answer customer questions and requests on our behalf, and if customers were still unhappy, they would be routed to an Engineer. 2. Choose your model Photo by Solen Feyissa on Unsplash You have two options: Train your model from scratch or use a pre-trained model and build on top of it. The latter would work in most cases unless you have a particular use case. Training your model from scratch will require massive computing power, significant engineering efforts, and costs, among other things. Now, the next question is, which pre-trained model should I choose? You can select a model based on your use case. 1B parameter model has basic knowledge and pattern matching. Use cases can be restaurant reviews. The 10B parameter model has excellent knowledge and can follow instructions like a food order chatbot. A 100B+ parameters model has rich world knowledge and complex reasoning. This can be used as a brainstorming partner. There are many models available, such as Llama and ChatGPT. Once you have a model in place, you can expand on the model. 3. Enhance the model as per your data Once you have a model in place, you can expand on the model. The LLM model is trained on generally available data. We want to train it on our data. Our model needs more context to provide answers. Let’s assume we want to build a restaurant chatbot that answers customer questions. The model does not know information particular to your restaurant. So, we want to provide the model some context. There are many ways we can achieve this. Let’s dive into some of them.  Prompt Engineering Prompt engineering involves augmenting the input prompt with more context during inference time. You provide context in your input quote itself. This is the easiest to do and has no enhancements. But this comes with its disadvantages. You cannot give a large context inside the prompt. There is a limit to the context prompt. Also, you cannot expect the user to always provide full context. The context might be extensive. This is a quick and easy solution, but it has several limitations. Here is a sample prompt engineering. “Classify this reviewI love the movieSentiment: PositiveClassify this reviewI hated the movie.Sentiment: NegativeClassify the movieThe ending was exciting” Reinforced Learning With Human Feedback (RLHF) RLHF Model RLHF is one of the most-used methods for integrating LLM into an application. You provide some contextual data for the model to learn from. Here is the flow it follows: The model takes an action from the action space and observes the state change in the environment as a result of that action. The reward model generated a reward ranking based on the output. The model updates its weight accordingly to maximize the reward and learns iteratively. For instance, in LLM, action is the next word that the LLM generates, and the action space is the dictionary of all possible words and vocabulary. The environment is the text context; the State is the current text in the context window. The above explanation is more like a textbook explanation. Let’s have a look at a real-life example. You want your chatbot to answer questions regarding your wiki documents. Now, you choose a pre-trained model like ChatGPT. Your wikis will be your context data. You can leverage the langchain library to perform RAG. You can Here is a sample code in Python from langchain.document_loaders import WikipediaLoader from langchain.text_splitter import RecursiveCharacterTextSplitter from langchain.embeddings import OpenAIEmbeddings from langchain.vectorstores import FAISS from langchain.chat_models import ChatOpenAI from langchain.chains import RetrievalQA import os # Set your OpenAI API key os.environ["OPENAI_API_KEY"] = "your-openai-key-here" # Step 1: Load Wikipedia documents query = "Alan Turing" wiki_loader = WikipediaLoader(query=query, load_max_docs=3) wiki_docs = wiki_loader.load() # Step 2: Split the text into manageable chunks splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=100) split_docs = splitter.split_documents(wiki_docs) # Step 3: Embed the chunks into vectors embeddings = OpenAIEmbeddings() vector_store = FAISS.from_documents(split_docs, embeddings) # Step 4: Create a retriever retriever = vector_store.as_retriever(search_type="similarity", search_kwargs={"k": 3}) # Step 5: Create a RetrievalQA chain llm = ChatOpenAI(temperature=0, model_name="gpt-3.5-turbo") qa_chain = RetrievalQA.from_chain_type( llm=llm, chain_type="stuff", # You can also try "map_reduce" or "refine" retriever=retriever, return_source_documents=True, ) # Step 6: Ask a question question = "What did Alan Turing contribute to computer science?" response = qa_chain(question) # Print the answer print("Answer:", response["result"]) print("\n--- Sources ---") for doc in response["source_documents"]: print(doc.metadata) 4. Evaluate your model Now, you have added RAG to your model. How do you check if your model is behaving correctly? This is not a code where you give some input parameters and receive a fixed output, which you can test against. Since this is a language-based communication, there can be multiple correct answers. But what you can know for sure is whether the answer is incorrect. There are many metrics you can test your model against.  Evaluate manually You can continually evaluate your model manually. For instance, we had integrated a Slack chatbot that was enhanced with RAG using our wikis and Jira. Once we added the chatbot to the Slack channel, we initially shadowed its responses. The clients could not view the responses. Once we gained confidence, we made the chatbot publicly visible to the clients. We evaluated its response manually. But this is a quick and vague approach. You cannot gain confidence from such manual testing. So, the solution is to test against some benchmark, such as ROUGE. Evaluate with ROUGE score.  ROUGE metrics are used for text summarization. Rouge metrics compare the generated summary with reference summaries using different ROUGE metrics. Rouge metrics evaluate the model using recall, precision, and F1 scores. ROUGE metrics come in various types, and poor completion can still result in a good score; hence, we refer to different ROUGE metrics. For some context, a unigram is a single word; a bigram is two words; and an n-gram is N words. ROUGE-1 Recall = Unigram matches/Unigram in referenceROUGE-1 Precision = Unigram matches/Unigram in generated outputROUGE-1 F1 = 2 * (Recall * Precision / (Recall + Precision))ROUGE-2 Recall = Bigram matches/bigram referenceROUGE-2 Precision = Bigram matches / Bigram in generated outputROUGE-2 F1 = 2 * (Recall * Precision / (Recall + Precision))ROUGE-L Recall = Longest common subsequence/Unigram in referenceROUGE-L Precision = Longest common subsequence/Unigram in outputROUGE-L F1 = 2 * (Recall * Precision / (Recall + Precision)) For example, Reference: “It is cold outside.”Generated output: “It is very cold outside.” ROUGE-1 Recall = 4/4 = 1.0ROUGE-1 Precision = 4/5 = 0.8ROUGE-1 F1 = 2 * 0.8/1.8 = 0.89ROUGE-2 Recall = 2/3 = 0.67ROUGE-2 Precision = 2/4 = 0.5ROUGE-2 F1 = 2 * 0.335/1.17 = 0.57ROUGE-L Recall = 2/4 = 0.5ROUGE-L Precision = 2/5 = 0.4ROUGE-L F1 = 2 * 0.335/1.17 = 0.44 Reduce hassle with the external benchmark The ROUGE Score is used to understand how model evaluation works. Other benchmarks exist, like the BLEU Score. However, we cannot practically build the dataset to evaluate our model. We can leverage external libraries to benchmark our models. The most commonly used are the GLUE Benchmark and SuperGLUE Benchmark.  5. Optimize and deploy your model This step might not be crucial, but reducing computing costs and getting faster results is always good. Once your model is ready, you can optimize it to improve performance and reduce memory requirements. We will touch on a few concepts that require more engineering efforts, knowledge, time, and costs. These concepts will help you get acquainted with some techniques. Quantization of the weights Models have parameters, internal variables within a model that are learned from data during training and whose values determine how the model makes predictions. 1 parameter usually requires 24 bytes of processor memory. So, if you choose 1B, parameters will require 24 GB of processor memory. Quantization converts the model weights from higher-precision floating-point numbers to lower-precision floating-point numbers for efficient storage. Changing the storage precision can significantly affect the number of bytes required to store a single value of the weight. The table below illustrates different precisions for storing weights. Pruning Pruning involves removing weights in a model that are less important and have little impact, such as weights equal to or close to zero. Some techniques of pruning are a. Full model retrainingb. PEFT like LoRAc. Post-training. Conclusion To conclude, you can choose a pre-trained model, such as ChatGPT or FLAN-T5, and build on top of it. Building your pre-trained model requires expertise, resources, time, and budget. You can fine-tune it as per your use case if needed. Then, you can use your LLM to power applications and tailor them to your application use case using techniques like RAG. You can evaluate your model against some benchmarks to see if it behaves correctly. You can then deploy your model.  The post A Step-By-Step Guide To Powering Your Application With LLMs appeared first on Towards Data Science.

Le conglomérat japonais des services IT vient de boucler un partenariat de taille avec la référence de l'IA, OpenAI. Effectif à compter du 1er...

8BitDo Retro Mechanical Keyboard - Xbox Edition $110.39 (was $120) See at Amazon See at Walmart (sold out) Retro R8 Mouse with Charging Dock - Xbox Edition $48 (was $60) See at Amazon See at Walmart ($60) 8BitDo's transparent green gaming keyboard and mouse are discounted at Amazon for a limited time. The Retro R8 Wireless Gaming Mouse is on sale for only $48 (was $60); it's the first discount for this budget-friendly wireless mouse since its January launch. The Retro 87 Wireless Mechanical Keyboard is up for grabs for $110 (was $120). These matching PC gaming peripherals are officially licensed by Xbox and pay homage to the popular green special-edition original Xbox. While you're checking these deals out, you should also take a look at the matching 8BitDo Ultimate 3-Mode Controller ahead of its May 15 release.8BitDo Xbox Edition Peripherals (Transparent Green)8BitDo Retro R8 Wireless Mouse - Xbox Edition -- $608BitDo Retro 87 Wireless Mechanical Keyboard - Xbox Edition -- $110 ($120)8BitDo Ultimate 3-Mode Controller - Xbox Edition -- $70 | Releases May 158BitDo Retro 87 Wireless Mechanical Keyboard - Xbox Edition$110.39 ($120) | PC, Mac, AndroidContinue Reading at GameSpot

The various iconic vehicles from the previous Subnautica games, such as the Cyclops submarine, won't be making a comeback in Subnautica 2, as confirmed by Design Lead Anthony Gallegos. Although relatively little is yet known about the ambitious and greatly-anticipated sequel, Subnautica 2is known to tell a new story taking place on a new planet and will focus on introducing new mechanics and features to deliver a fresh experience for players.

Each week on Polygon, we round up the most notable new releases to streaming and VOD, highlighting the biggest and best new movies for you to watch at home. This week, Havoc, the long-awaited new film from The Raid director Gareth Evans and starring Tom Hardy, bursts its way onto Netflix. There are tons of other new releases on Netflix alone, including a Japanese action thriller perfect for fans of Unstoppable or Speed. There’s plenty to watch on other streaming services, too, like Babygirl starring Nicole Kidman on Max, The Return on Paramount Plus, My Hero Academia: You’re Next on Crunchyroll, and the sci-fi horror thriller Ash on Shudder. Last but not least, this week’s VOD releases include No Other Land, the Oscar-winning documentary that’s currently available to rent for a limited time only. Here’s everything new that’s available to watch this weekend! New on Netflix Havoc Where to watch: Available to stream on Netflix Genre: Action thrillerRun time: 1h 45mDirector: Gareth EvansCast: Tom Hardy, Jessie Mei Li, Timothy Olyphant Director Gareth Evans (The Raid, Gangs of London) is back with his first new film in nearly seven years and his first action thriller since 2014’s The Raid 2. Tom Hardy stars as Walker, a police detective who is sent on a perilous mission to rescue the estranged son of a powerful politician after a drug deal gone bad. Hunted throughout the city, Walker will have to use every ounce of his cunning and every bullet to his name in order to make it out alive. Bullet Train Explosion Where to watch: Available to stream on Netflix Genre: Disaster actionRun time: 2h 14mDirector: Shinji HiguchiCast: Tsuyoshi Kusanagi, Kanata Hosoda, Non From Shin Godzilla director Shinji Higuchi, Bullet Train Explosion follows a high-speed Japanese bullet train that gets hijacked by a mysterious terrorist. If the train goes below 100 kilometers per hour, a bomb will go off, killing everyone on board — unless the 100 billion yen ransom is paid. Will the ransom be paid? Will the government let the train explode? Can the conductor somehow get the innocent passengers to safety? Will they enact an extravagant rescue mission involving two trains? I have no idea, but man, the situation is tense. The movie is a sequel to the 1975 classic The Bullet Train, which itself was a big influence on Speed. Pangolin: Kulu’s Journey Where to watch: Available to stream on Netflix Genre: DocumentaryRun time: 1h 28mDirector: Pippa Ehrlich Pangolins are some of the coolest and most adorable creatures on the planet. Just look at them; they look like real-life Pokémon. Unfortunately, this very quality has made them the target of poachers and one of the most trafficked animal species in the world. This documentary follows the story of a man who finds new purpose after he rescues a baby pangolin, nursing it back to health and working tirelessly to rehabilitate it to return to the wild. New on Max Babygirl Where to watch: Available to stream on Max Genre: Erotic thrillerRun time: 1h 54mDirector: Halina ReijnCast: Nicole Kidman, Harris Dickinson, Antonio Banderas In Babygirl, Romy, a stressed and sexually unsatisfied CEO (Nicole Kidman) starts a dominant-submissive affair with her intern Samuel (Harris Dickinson). As the affair intensifies, Samuel starts interjecting himself more into Romy’s life, threatening to upset both her career and her home life. The erotic thriller generated a lot of buzz when it came out, with comparisons to Fifty Shades of Grey and Secretary and praise for Kidman’s vulnerable performance. New on Paramount Plus The Return Where to watch: Available to stream on Paramount Plus Genre: DramaRun time: 1h 56mDirector: Uberto PasoliniCast: Ralph Fiennes, Juliette Binoche, Charlie Plummer Uberto Pasolini’s dramatization of Homer’s Odyssey stars Ralph Fiennes as Odysseus, the king of Ithaca who washes ashore in his homeland 20 years after departing for the Trojan War. Scarred mentally and physically by his decades-long ordeal to return home, Odysseus must muster all of his cunning and strength to reclaim his household from a band of squatting suitors vying to marry his wife, Penelope (Juliette Binoche). New on Peacock Last Breath Where to watch: Available to stream on Peacock Genre: ThrillerRun time: 1h 33mDirector: Alex ParkinsonCast: Woody Harrelson, Simu Liu, Finn Cole Director Alex Parkinson returns with a survival thriller remake of his 2019 documentary Last Breath. Based on a true story, the film centers on a crew of seasoned deep-sea divers as they race against time to save their teammate after his umbilical cable is severed, leaving him stranded 330 feet below sea level with no light, no heat, and a dwindling amount of oxygen. New on Kanopy Banned Together Where to watch: Available to stream on Kanopy Genre: DocumentaryRun time: 1h 33mDirectors: Kate Way, Tom WigginCast: Juno Dawson, Jodi Picoult, Isabella Troy Brazoban Three teenage students and their adult allies band together in order to fight a sudden book ban that pulled 97 books from their school libraries. Banned Together showcases their journey as they attend protests, make speeches, and head from local school board meetings to the United States Congress in order to present their case and fight for the right to read. The documentary also dives into Moms for Liberty, the conservative organization responsible for a lot of the challenged books in American libraries. New on Crunchyroll My Hero Academia: You’re Next Where to watch: Available to stream on Crunchyroll Genre: Superhero actionRun time: 1h 50mDirector: Tensai OkamuraCast: Daiki Yamashita, Nobuhiko Okamoto, Yuki Kaji The newest My Hero Academia movie takes place right after the U.A. Traitor arc in the manga. A villain is impersonating All Might, so Izuku Midoriya and the rest of his friends in Class 1-A team up to stop the baddie! Along the way, the gang meets some new characters with some cool new quirks… and they also get embroiled in a mafia plot! Ain’t that just the way of things in the My Hero world? New on Shudder Ash Where to watch: Available to stream on Shudder Genre: Sci-fi HorrorRun time: 1h 35mDirector: Flying LotusCast: Eiza González, Aaron Paul, Iko Uwais Flying Lotus is back in the director’s chair with a new sci-fi horror thriller. Set on an alien world light-years from Earth, Ash stars Eiza González (3 Body Problem) as Riya, a colonist who wakes up to discover her crewmates have been viciously slaughtered. With no memory of what occured, she’ll have to decide whether or not she can trust Brio (Aaron Paul), the only other surviving member of the expedition, if she’s to have any hope of unraveling the mysteries of his hostile world and escaping alive. Polygon interviewed Flying Lotus about the production behind film, as well as González and Paul about their experience on set. Fréwaka Where to watch: Available to stream on Shudder Genre: HorrorRun time: 1h 43mDirector: Aislinn ClarkeCast: Clare Monnelly, Bríd Ní Neachtain, Aleksandra Bystrzhitskaya Fans of Midsommar, Enys Men, and The Ritual will love this new folk horror thriller from director Aislinn Clarke. Haunted by a recent loss, home care worker Shoo (Clare Monnelly) is sent to a remote village in the Irish countryside to look after an agoraphobic woman who fears she’ll be abducted by a supernatural force. While at first she doubts her new charge, Shoo slowly but surely grows to suspect that something is not right about this village. New to rent No Other Land Where to watch: Available to rent on supportmasaferyatta.com Genre: DocumentaryRun time: 1h 32mDirectors: Basel Adra, Hamdan Ballal, Yuval Abraham, Rachel Szor Recorded between 2019 and 2023, this documentary conceived by a Palestinian-Israeli collective of activists follows the devastation of a Palestinian community in the occupied West Bank following the declaration of an Israeli “firing zone” on their land. Despite winning Best Documentary at his year’s Academy Awards, No Other Land has not been able to find a distributor in the United States, with the only way to see the movie being limited screenings at select theaters across the country. That is, until now. Last week, the creators have made No Other Land available to rent for a limited time in order to raise money for Masafer Yatta, the community depicted in the film. If you’re interested in watching the film, act fast — you have until May 9 in order to rent it. Locked Where to watch: Available to rent on Amazon, Apple, and Vudu Genre: Action thrillerRun time: 1h 35mDirector: David YaroveskyCast: Bill Skarsgård, Anthony Hopkins When Eddie (Bill Skarsgård), a young car thief, breaks into a luxury SUV he happens upon in a secluded parking lot, he gets more than he bargained for when he discovers he’s been trapped inside. When the vehicle’s owner, William (Anthony Hopkins), puts him through a series of tests with the potential of escape, Eddie finds himself ensnared in a hell of his own making. Produced by Sam Raimi, Locked looks like a fucked-up version of Carpool Karaoke by way of Saw, and if that doesn’t sound like a fun time, I don’t know what does.