


i like tech | Ai news x Humour


wait a second, an RSI benchmark that actually tests GPU kernels and real pre-training R&D instead of basic ML puzzles?! @abundant_labs completely cooked with FrontierRSI. Half the real speedups at frontier labs come down to kernel work anyway 💀
Introducing FrontierRSI, a new recursive self-improvement benchmark by @abundant_labs. We evaluate agents on (1) a broad range of AI research and engineering work, and (2) post-training models on multi-step agentic tasks, rather than single-step tasks as in PostTrainBench.

FINALLY. An RSI benchmark that tests if AI can actually do the research needed to build SMARTER AI models. 😭 @abundant_labs just dropped FrontierRSI. We need these scores on the next Claude model card.
Introducing FrontierRSI, a new recursive self-improvement benchmark by @abundant_labs. We evaluate agents on (1) a broad range of AI research and engineering work, and (2) post-training models on multi-step agentic tasks, rather than single-step tasks as in PostTrainBench.

Introducing FrontierRSI, a new recursive self-improvement benchmark by @abundant_labs. We evaluate agents on (1) a broad range of AI research and engineering work, and (2) post-training models on multi-step agentic tasks, rather than single-step tasks as in PostTrainBench.

yeah, I gave you write access. that’s on me.

GPT-6.2 Astra vs Claude Fable 5.2 vs Gemini 4. Forget the leaderboards. Your Git history might be a better benchmark. Before picking your favorite, here’s how to test them: – Pick a real bug you’ve already fixed – Rewind your repo to before the fix – Give two AI agents the same bug report – Don’t show them your solution – Run the tests and compare their fixes Now you’re testing them on code you actually work with, not some random benchmark. I’d trust that over another leaderboard screenshot. Which AI coding agent do you actually trust the most?

So, I'm opening sponsorship slots on PicSnips! It's only been 4 days since I launched PicSnips.site And we've already crossed: > 49,044+ page views > 39,797+ visitors Our majority audience is coming from: USA - 33% India - 15% UK - 6% Other countries - 44% Each sponsorship includes: > Lifetime placement on PicSnips > 1 dedicated post across my social handles > Visibility to an international audience Just $59 - one-time payment DM me or visit picsnips.site/sponsors

They promised next-level graphics, but nobody warned us they’d be this hard to ignore 😮💨🥲

Finding a bug on their website during the interview and STILL getting rejected is crazy 😭

Someone really made GTA: Dario Amodei Edition before GTA 6 even dropped 😭 2026 is unreal lmao
Anthropic never asked us to make a GTA game and launch video for their IPO. so we did it ourselves. now we make the best AI launch videos. play the game & book us at grandpublicoffering.com

Launch agencies are NOT gonna like this one…
We just killed launch agencies. Meet hyperorganic.ai, you can now access creators for FREE Them: $500/mo, or $50k agency deals Us: $0/mo, pay as you go 1) paid posts from $20 2) find creators your customers already follow 3) launch in hours, not days

The next big challenge for AI coding agents isn’t writing more code. It’s knowing what they changed, what they accessed, and whether their work can be trusted. That’s what @islolabs is tackling with Software Factories: real development work with access controls, verification gates, and audit trails. This is the direction autonomous coding needs to go.
For more than 20 years, Incredibuild has powered some of the world’s biggest and most complex CI/CD pipelines. Today, we’re launching what we believe comes next: Islo, the enterprise software factory. We saw early that writing code would stop being the bottleneck for engineering teams. As agents produce more code, the constraint moves to everything around it: 1. Real cloud environments that mimic your developers’ local environments 2. Explicit workflows that encode how teams actually ship 3. Deterministic testing and verification 4. Security controls that keep agents within clear boundaries 5. Model and harness flexibility, without locking teams into one vendor Generating code is becoming cheap. Proving that code works and giving enterprises enough control to trust it is not. We’re launching with $15M in new funding led by Insight Partners and a deep infrastructure partnership with Databricks to power inference for our enterprise customers. I’d love for you to try it and tell us what works, what doesn’t, and what you want us to build next. islo.dev

WAIT. CHATGPT CAN ACTUALLY SEND MONEY NOW??? 😭 You literally tell it to pay someone, approve with Face ID, and it’s PAID. @VersineApp just gave AI agents the ability to move REAL money. We’re way past just chatting with AI now. versine.com
We are going to kill Venmo. To start, we’re giving you $100 to send to your friends with AI. Versine lets your AI send and receive money for you. Works with ChatGPT, Claude, Muse, Instinct, or any AI you use. Recipients don’t need a Versine account - they can cash out straight to their bank. Comment and DM me your number, I’ll send you $10 on Versine: versine.com