Sam V

**Opinions are my own and do not reflect those of any organization**

Research Scientist in AI4Science; LLMs and agentic AI.

[Sam V @SamVeras]

@vansh22b Most user don’t spend $200 monthly and most users don’t run over their usage weekly. You’re calculations are overestimating on all fronts.

[Sam V @SamVeras]

@thsottiaux Sr Reset Engineer?

[Sam V @SamVeras]

@PrashantByte09 LLMs are non-deterministic and AI tools are not the same harness all around.

[Sam V @SamVeras]

@araseb_ I wonder why no one talks about Kiro—the CLI not IDE.

[Sam V @SamVeras]

@jun_song 😆😆😆

[Sam V @SamVeras]

@Im_IrushiK This is not new. It’s been like this for years now.

[Sam V @SamVeras]

I disagree with @elonmusk on this one. Problem is not only generalization, but also the creative part of surgery. What happens when the robot is faced with a scenario it wasn’t expecting? And to be clear, I think robots would eventually be able to work well within medical sciences and other sciences, but high-risk fields such as surgery and family medicine will require more time than 3-5 years. Theoretical capabilities do not translate to deployable capabilities and Elon knows this.

[Vivek Sen @Vivek4real_]

ELON MUSK'S JAW-DROPPING PREDICTION:

"DON'T GO INTO MEDICAL SCHOOL."

ELON MUSK: "YES. POINTLESS."

IN 3 YEARS (2029), OPTIMUS ROBOTS WILL BE BETTER SURGEONS THAN ANY HUMAN ON EARTH — AT SCALE.

BY 4–5 YEARS? NOT EVEN CLOSE.

THE BEST MEDICINE IN THE WORLD WILL BE FREE — BETTER THAN WHAT THE PRESIDENT GETS TODAY.

THIS IS WILD.

[Sam V @SamVeras]

@kimmonismus I can agree to this and throw Opus 5 in the mixed. Yet I don’t think staying with sonnet 4.6 was feasible either. So I guess I would have preferred Anthropic had spent more time “fixing” sonnet 5 than releasing Opus 5.

[Sam V @SamVeras]

@Milajoy I would flip the order if given the option.

[Sam V @SamVeras]

@ClaudeDevs Nothing a usage limit reset can’t help us move pass from.

[Sam V @SamVeras]

Is there any data/paper to back these claims? And I don’t mean just numbers in the vacuum. I mean task benchmarks quantifying cost per task on both context window (258k vs 1M) where once compact say 3 times and the other just fills its own up to 750k. I’d be curious to know the trade-offs.

[Sam V @SamVeras]

@rough__sea Who’s IDE? Who’s JSON?

[Sam V @SamVeras]

@ben_sage Funny you ask. Just went back to it for the fourth time; cli looking a bit cleaner now. Gemini cli that is.

[Sam V @SamVeras]

@ericlim @ClaudeDevs do the thing.

[Sam V @SamVeras]

@gdb @thsottiaux does this mean higher usage limits?

[Sam V @SamVeras]

@gdb This is awesome! @ClaudeDevs you should definitely do a usage limit reset. It would allow us to stop reading news and keep building more. Can we get a reset? 😃

[Sam V @SamVeras]

@lightsilver323 http://localhost:8000

[Sam V @SamVeras]

@ClaudeDevs can we get a reset?

[Theo - t3.gg @theo]

Claude Code has been mostly down for half an hour. Really inconvenient, but I’ll forgive them if we get a reset

[Sam V @SamVeras]

@BenjaminDEKR I guess it’s time to touch grass and take some sunlight.

[Sam V @SamVeras]

@thsottiaux Good to know! I recently launched a task with my own agentic framework that guides GPT-sol and other models to run for days if necessary to accomplish their goals. That reset was appreciated—thanks.