Andranik Sahakyan

AI Engineer | Biohacker

[a16z @a16z]

Against all odds.

Congrats @elonmusk and @SpaceX.

[Andranik Sahakyan @antonplex]

@TheAhmadOsman Does the Camofox setup for browser automation work better than agent browser with local/open source models?

[Andranik Sahakyan @antonplex]

Finally got FSD 14.3.2! Still waiting on the Spring Update

[Andranik Sahakyan @antonplex]

@signulll Evolutionary version of backend for frontend, same pattern with agents using intent-based MCP tools rather than chaining multiple lower level REST API calls.

[Andranik Sahakyan @antonplex]

@BenjaminDEKR It was previously only on the mobile app

[Andranik Sahakyan @antonplex]

@CoachDanGo 2 eggs
turkey bacon
avocado toast with EVOO

[Andranik Sahakyan @antonplex]

@StockMKTNewz @Gemini card, getting BTC back on every transaction

[Andranik Sahakyan @antonplex]

@readswithravi The Bitcoin Standard by @saifedean

[Andranik Sahakyan @antonplex]

@KobeissiLetter What happens when stocks are fully tokenized allowing for 24/7 trading?

[Andranik Sahakyan @antonplex]

@HHorsley We don’t have a definition of AGI that is universally agreed on. We have markets for which model will be at the top of a certain leaderboard (e.g. top model on LMArena by end of month).

[Andranik Sahakyan @antonplex]

@ZachWarunek https://t.co/JVoliqoXLO

[Elon Musk @elonmusk]

Congrats to the @Tesla_AI chip design team on taping out AI5!

AI6, Dojo3 & other exciting chips in work.

[Andranik Sahakyan @antonplex]

@Yuchenj_UW Try @PrimeMeridianAI, they got 100% on TaxCalcBench. Did mine end to end for free, got my refund within a week.

[Andranik Sahakyan @antonplex]

@Vtrivedy10 Is there an easier way of disabling the general purpose subagent for deepagents other than using create_agent and manually constructing the whole stack?

[Andranik Sahakyan @antonplex]

AI will never write a witty reply to this post using self-referential paradoxes.

(The above reply was written by Grok)

[Mark Kretschmann @mark_k]

A statement starting with “AI will never...” is almost always doomed to be wrong.

[Andranik Sahakyan @antonplex]

Maybe we need a Turing test benchmark for AI image/video models.

[gabriel @gabriel1]

it's extremely hard to imagine what 100% perfect and solved image or video generation looks like

every new model launch "100% solves video/image" but then we continue to say that for 10 more releases

[Andranik Sahakyan @antonplex]

Optimus on duty at the @Tesla Diner!

[Andranik Sahakyan @antonplex]

@farzyness You should give @PrimeMeridianAI a try, gets 100% on TaxCalcBench.

[Andranik Sahakyan @antonplex]

@Vtrivedy10 Evaluating instruction-following for compliance-heavy use cases (SOC 2)

[Andranik Sahakyan @antonplex]

@andrewchen AI native products expose MCPs or CLIs optimized for use by agents. You can tell the difference even by seeing what APIs they expose (querying/filtering mechanisms, token efficiency, etc).