AI

Our latest model, Claude Opus 4.7, is now generally available.

Opus 4.7 is a notable improvement on Opus 4.6 in advanced software engineering, with particular gains on the most difficult tasks. Users report being able to hand off their hardest coding work—the kind that previously needed close supervision—to Opus 4.7 with confidence. Opus 4.7 handles complex, long-running tasks with rigor and consistency, pays precise attention to instructions, and devises ways to verify its own outputs before reporting back.

The model also has substantially better vision: it can see images in greater resolution. It’s more tasteful and creative when completing professional tasks, producing higher-quality interfaces, slides, and docs. And—although it is less broadly capable than our most powerful model, Claude Mythos Preview—it shows better results than Opus 4.6 across a range of benchmarks:

cross-posted from: https://lemmy.ml/post/45997309

April 14, 2026

Most Americans using AI tools for health purposes say they want immediate answers. In some cases, it helps them evaluate what kind of medical attention they need.

“It’ll let me know if something’s serious or not,” Davis said of ChatGPT, which she typically consults before scheduling medical appointments.

The Gallup survey found about 7 in 10 U.S. adults who have used AI for health research in the past 30 days say they wanted quick answers, additional information or were simply curious. Majorities used it for research before seeing a doctor or after an appointment.

cross-posted from: https://lemmy.world/post/45587578

Does the world need more billionaires? Guess we will soon find out, as they multiplying now, and not even naturally!

I've been playing with Gemma4, and in one instance it took me something like half an hour to have it acknowledge a statement I made, denying the contents of an official Google page I asked it to search and parse. It lied, and at some point suggested I was hallucinating! If you install it and try it, put it to the test, hard.

cross-posted from: https://lemmy.world/post/45435884

"the company admitted it likely won’t be able to keep up with competing models."

"As such, the announcement is a bit of an enigma: if it can’t keep up with the competition, why release it at all? There’s a good change Meta is just trying to get its foot in the door — or a “seat at the big kid’s table,” as Wired put it. The company has struggled to stay relevant in a rapidly changing landscape" "Meta’s preceding Llama open source models largely failed to catch on, with a major controversy last year finding that Meta may have faked benchmark results to make its Llama 4 model seem more capable than it actually was."

cross-posted from: https://lemmy.world/post/45435064

So I decided to (literally) TEST them!

The result?

...well see the wonders of Telegram AI for yourselves 🤣 🤣 🤣

Chinese AI company Z.ai has launched GLM-5.1, an open-source coding model it says is built for agentic software engineering. The release comes as AI vendors move beyond autocomplete-style coding tools toward systems that can handle software tasks over longer periods with less human input.

Z.ai said GLM-5.1 can sustain performance over hundreds of iterations, an ability it argues sets it apart from models that lose effectiveness in longer sessions.

As one example, the company said GLM-5.1 improved a vector database optimization task over more than 600 iterations and 6,000 tool calls, reaching 21,500 queries per second, about six times the best result achieved in a single 50-turn session.

In a research note, Z.ai said GLM-5.1 outperformed its predecessor, GLM-5, on several software engineering benchmarks and showed particular strength in repo generation, terminal-based problem solving, and repeated code optimization. The company said the model scored 58.4 on SWE-Bench Pro, compared with 55.1 for GLM-5, and above the scores it listed for OpenAI’s GPT-5.4, Anthropic’s Opus 4.6, and Google’s Gemini 3.1 Pro on that benchmark.

Can someone who knows more about AI agents explain this to me?

Axios quotes an Anthropic spokesperson re: the Claude leaks, who says that this new capability is coming: "A "persistent assistant" running in background mode that lets Claude Code keep working even when a user is idle. "

Isn't that what they claim "agents" are doing already? Or am I missing something? I would hate to learn that "agents" are just a marketing term and they don't actually do the thing these companies claim... ;)

Presearch Residency Program

flagship program for enabling the best research talent across all disciplines to shape the future of AI. exceptional researchers, engineers, and analysts from any discipline to help advance the impact of AI research. If you’re passionate about shaping agentic systems, developing human–AI interaction, and pushing the boundaries of interdisciplinary work, we encourage you to apply.

Program length: 3 ay tam zamanlı ofiste 3 months, full-time commitment

Location In-person at our San Francisco or Palo Alto office.

I could not fit everything i wanted to ask in the title, and the title probably needs some explanation. So to preface, I dont use AI code much, I have several projects which I taken effort to have 0% AI code, but I was wondering, because I do have certain tools, e.g a timetable editor supposed to make a timetable in a specific format i liked, or something that generates or give me a links of docs in a specific way, that I had generated, I also had projects which I tried using it for snippets, boilerplate and explanations. But I was wondering, where would be like, the best balance between human and AI code be, or rather how would you reclaim it in a sense, or improve your learning while also being efficient or also be knowledgeable enough about your code that you are a suitable person to maintain it.

Like, is it the second you read such code and try to understand it, you have the understanding but you completely lost most learning? Also would you potentially be locked into bad design choices by AI. What if you just looked at the inputs of the programs or the outputs and tried to recreate from that. Or what if you asked it what libraries to use, and you use those, would you have deprived yourself significantly of some level of understanding by having the design and architecture handing too you, even if you do the rest of the coding, and finally what if you memorized enough of the code to the point you can recreate it, and maybe apply it in different contexts a bit. Also would it be worth sharing if i do understand such code that was made by the AI, or would i be contributing to Slop. Basically, can you use it, but still at the end of the day be suitable to maintaining that code and etc.

Sorry if thats alot, anyways, curious to hear what people think, this could be useful for me in the future.

Ethiopia is positioning itself as a leading artificial intelligence player in Africa, with local innovation—including components of Sophia the robot developed in Addis Ababa—and policy initiatives such as a new AI institute and a forthcoming national AI strategy. Guests on the episode include Hruy Tsegaye, co-founder of iCog Labs, Abdullahi Tsanni of MIT Tech Review, and Dr. Taye Girma, Deputy Director General of the Ethiopian Artificial Intelligence Institute. Dr. Girma notes that the institute is working on how AI can boost sectors including health, finance, transportation, and agriculture, while discussions also address ethical challenges and the development of AI tools for local languages such as Amharic.

Vectors are the fundamental way AI models understand and process information. Small vectors describe simple attributes, such as a point in a graph, while “high-dimensional” vectors capture complex information such as the features of an image, the meaning of a word, or the properties of a dataset. High-dimensional vectors are incredibly powerful, but they also consume vast amounts of memory, leading to bottlenecks in the key-value cache, a high-speed "digital cheat sheet" that stores frequently used information under simple labels so a computer can retrieve it instantly without having to search through a slow, massive database.

Vector quantization is a powerful, classical data compression technique that reduces the size of high-dimensional vectors...

archive.org

ghostarchive.org