Arthur Welf 🇺🇦🇮🇹🇪🇺

[Rep. Jack Kimble @RepJackKimble]

We can’t tolerate Ukraine interference in the midterm elections.

[Arthur Welf 🇺🇦🇮🇹🇪🇺 @welf]

@thsottiaux restarting mcp servers without restarting the app/cli

[WarMonitor🇺🇦🇬🇧 @WarMonitor3]

Incredible video shows a Russian drone attempting to destroy a Ukrainian ground robot but being disabled when the robot deploys an anti drone net and catches it.

We literally have robots fighting each other on Ukraine’s frontline.

[☄︎ @0xca0a]

We are releasing → https://t.co/9bfbbdJZip today: AI driven embeddable CAD. If you ever enjoyed my OSS projects (zustand, react-spring, r3f, pmndrs), these all came out of CC. 🍻
A small 🇨🇭 team has been working on CAD for so many years, release seemed always out of reach. This is a huge milestone for us and i'd appreciate a RT. 🙏

[Arthur Welf 🇺🇦🇮🇹🇪🇺 @welf]

@thsottiaux @thsottiaux what is the procedure to unblock OpenAI models to work on dual-use/military products?

[klöss @kloss_xyz]

I handed Claude Design one of my brand's design systems to see how good a video editor, motion designer, and sound mixer it was.

Opus 5.5 also wrote the Suno v6 prompts: a 150 bpm song and the 10 sound effects used here.

Suno gave me 2 takes of each and I picked my favorites by ear. Claude can't hear, so it measured tempo, loudness, and every drop to cut the sound.

Then I passed it to Claude Code to render frame by frame and stitch the mp4, since the render was so big Claude Design kept crashing.

Here's the end result. Insane 1st render.

[Arthur Welf 🇺🇦🇮🇹🇪🇺 @welf]

@kloss_xyz Try to add your agents code intelligence tools: https://t.co/jiZPckpgmt (local MCP server)

They are providing the same productivity gain for your agents as human developers got moving from text notepad to IDEs. It is especially visible in large codebases.

[Taelin @VictorTaelin]

Bend 2 is here!

It is a new programming language that blocks AI mistakes via *proof checking* - the same technique big AI labs used to solve open math problems, like Navier-Stokes.

It is also very fast, and runs on GPUs.

Watch the video. Link in the comments.

[Arthur Welf 🇺🇦🇮🇹🇪🇺 @welf]

@unclebobmartin Are you planning to add to your harness code intelligence tools for your agents (actually, the tools every human developer has in its IDE)? I have built them for my agents and would be glad to have a feedback from you: https://t.co/jiZPckoIwV

[Arthur Welf 🇺🇦🇮🇹🇪🇺 @welf]

@diodo_e Hi from Rome

[Arthur Welf 🇺🇦🇮🇹🇪🇺 @welf]

@diodo_e Code-intelligence tools for your agents: https://t.co/jiZPckoIwV

[Generalist @GeneralistAI]

Introducing GEN-1.5, a one-shot learner.

It can learn new tasks in a few seconds. Show it what to do, and it generalizes.

This capability emerged from pretraining on physical data at scale, as a step towards our mission of building general intelligence for the physical world.

[Arthur Welf 🇺🇦🇮🇹🇪🇺 @welf]

@formigarafa @unclebobmartin @wmksidebottom we should verify what the code does, not the agent.

[Arthur Welf 🇺🇦🇮🇹🇪🇺 @welf]

@thsottiaux It tends to over-engineer solutions. When asked not to do it, tends to over-react to this rule and cuts the necessary features defined in the plan as “over-engineering”.

[Arthur Welf 🇺🇦🇮🇹🇪🇺 @welf]

@thsottiaux Let me create a PR to fix the bug in Codex that affects MCP servers and causes the `Transport Closed` issue: https://t.co/wMI7Su0R7Z

The fix is ready and tested but I can't create a PR that will resolve many of your open issues.

[Arthur Welf 🇺🇦🇮🇹🇪🇺 @welf]

@galdawave @karrisaarinen The high cache/other tokens ratio can witness the problem with tooling/tool use. Instead of batching tool calls, it calls sequentially and every tool output appends to previous cached results. Too many small steps -> high cache usage.

[Arthur Welf 🇺🇦🇮🇹🇪🇺 @welf]

@thsottiaux @thsottiaux You have many dozens issues with multiple MCP servers. I've fixed it but can't open a PR: https://t.co/wMI7Su0R7Z
Let me open the PR and you'll be able to close a whole class of issues

[Arthur Welf 🇺🇦🇮🇹🇪🇺 @welf]

@unclebobmartin @ori_pomerantz I use the same strategy. Strong type system, a lot of lints, a lot of different kinds of tests… It solves the problem of the local code quality but not at the global level. How are you dealing with violations of DRY and other global principles?

[Arthur Welf 🇺🇦🇮🇹🇪🇺 @welf]

@ProductHunt @OpenAI @OpenAIDevs What if only a small (final) part of the large product was made with GPT-5.6?

[Arthur Welf 🇺🇦🇮🇹🇪🇺 @welf]

I was a paying customer of Anthropic since November 2024 (Sonnet 3.5, pre-Claude Code era). Now I’m considering to cancel my Max x20 subscription. Comparing to Codex it is so bad at following the instructions! I’m really tired of writing hooks. Codex is awesome in it!