The End of SWE-Bench Verified (2024-2026) https://
latent.space/p/swe-bench-de
ad
… Today @OpenAIDevs is announcing the voluntary deprecation of SWE-Bench Verified! We're releasing a podcast + analysis in today's post. Saturation of SWE-Bench has been a community hot topic for over a year –
CODE
-
SWE-Bench Verified Deprecation: End of Major AI Benchmark
By
–
-

Understanding SiLU/GELU Activation Performance Loss in RL Networks
By
–
I always lost performance when I tried to use silu/gelu activations in my RL value networks, and I finally understand why. If the pre-activation values are small, the smooth curve through zero is basically a linear activation, destroying the representation power of the network. You need a batch/layer/rms norm on the preactivations to put them in the range the smooth activations are designed for. Internal norms generally hurt performance on our RL tasks, but combining them with a smooth activation at least works basically as well as a raw relu (but slower). So, not actually a win, but the lightbulb of understanding was good!
→ View original post on X — @id_aa_carmack, 2026-02-23 16:54 UTC
-
Customizing AI Assistant Personality and Instructions
By
–
not sure if you’ve tried this but you can change the personality w/ “/personality” – there’s two options “friendly” and “pragmatic” (default) in addition to that you can add custom instructions that you want codex to follow and it nudges its responses towards that
-
Indian Government NSO MCP Server Launch and Data Pipeline Impact
By
–
When I heard of Indian Govt (NSO) launching a MCP Server, didn't quite understand the impact. Watching this video, gives me context: "We hold the tap for key data pipelines – like that of NSO." Thank you @Krishnaik06 for simple explanation. piped.video/watch?v=-UQ6OZyw…
→ View original post on X — @krishnaik06, 2026-02-23 07:04 UTC
-

Becoming a Dev in 2026: Mastering AI Agentic Systems
By
–
If you're a developer in France and wondering what to learn in 2026, this post is for you 👇 I don't communicate much in French, but if I can help French speakers make the right choices about what to learn, I will. Let me introduce myself: I'm Brivael, tech co-founder of @argildotai, I went through Y Combinator in 2024 and raised 5 million euros. We're currently a team of 11 people based in Paris and we're building what will be seen in the future as a kind of AGI / Agent to produce videos for social media. Today we have tens of thousands of users, we're focused on video generation with AI avatars, but we're going to expand to much more than that very soon. If you want to ride the AI wave, it's still possible. But you have to be honest about what it takes. To paraphrase @naval: there will no longer be demand for average people. It's brutal but it's reality. If you don't find a way to be in the top 5% of what you do, you'll be unemployable. Not in 10 years. Soon. At Argil we have 7 tech people internally. And we're automating almost the entire software construction process. Our days increasingly boil down to one thing: designing agentic systems, validating architecture, orchestrating AI agents to do the rest. Classic code? Agents already do it. What has value today is the ability to think in systems, to understand how to orchestrate models together, to know when an agent should decide and when it should delegate. And that's where many get it wrong: even if AI produces 100% of the code, without mastery of software engineering, architecture and systems as a whole, it's almost impossible to [Translated from EN to English]
→ View original post on X — @flashtweet, 2026-02-22 23:08 UTC
-

Google AI Studio supports React and Next.js
By
–
ICYMI: Google AI Studio now supports React and Next.js frameworks on top of Angular. A preparation for something bigger. "XR Blocks" is in development
-

29k GitHub Contributions: Building and Shipping Consistently
By
–
29,023 contributions in the last year. Building quietly. Shipping consistently.
Compounding daily. How did I do? GitHub: https://
github.com/MontrealAI -
Databricks AI Functions Extract Critical Data From PDFs
By
–
Databricks AI Functions and the Information Extraction Agent Brick help teams unlock critical information trapped in PDFs and documents.
— Databricks (@databricks) 22 février 2026
By combining built-in AI with domain-specific agents, teams can quickly extract key data, validate it with experts, and turn unstructured… pic.twitter.com/PkNUsM0aniDatabricks AI Functions and the Information Extraction Agent Brick help teams unlock critical information trapped in PDFs and documents. By combining built-in AI with domain-specific agents, teams can quickly extract key data, validate it with experts, and turn unstructured
-

Scikit-Learn Classification Techniques for Machine Learning
By
–
Scikit-Learn for Classification in Machine Learning. @kdnuggets #BigData #Analytics #DataScience #AI #MachineLearning #IoT #IIoT #Python #RStats #TensorFlow #JavaScript #ReactJS #CloudComputing #Serverless #Linux #Programming #Coding #100DaysofCode https://
geni.us/Scikit-Classif -

Essential Mathematics for Data Science and Big Data Analytics
By
–
Math for #DataScience! #BigData #Analytics #DataScience #IoT #IIoT #PyTorch #Python #RStats #TensorFlow #Java #JavaScript #ReactJS #GoLang #CloudComputing #Serverless #DataScientist #Linux #Books #Programming #Mathematics #Coding #100DaysofCode https://
geni.us/Math-DSci