Aditya Jhalani

17
@TrinitySchoolNY '28
nyc

[Aditya Jhalani @adityajhalani43]

@lmoroney being able to load GGUF straight through transformers makes local models way less annoying to actually use

[Aditya Jhalani @adityajhalani43]

@kimmonismus The 15% hallucination rate comes with lower answer accuracy. That tradeoff gets lost if you only look at the overall score.

[Aditya Jhalani @adityajhalani43]

@0xAruAziki Prompt injection in the evidence might be the bigger problem. A jury of models doesn’t help much if they all read the same poisoned input.

[Aditya Jhalani @adityajhalani43]

@a16z token usage gets a lot less useful as a growth metric when agents consume 5x as much as people

[Aditya Jhalani @adityajhalani43]

@ideogram_ai the 10th edit matters more than the first. keeping colors and small details stable that long would make this way more useful.

[Aditya Jhalani @adityajhalani43]

@nikhilx22 @X I'm building AI agents too. Would be good to connect.

[Aditya Jhalani @adityajhalani43]

@KhandharRushabh @X i've been building with AI agents and music ML lately. down to connect

[Aditya Jhalani @adityajhalani43]

@jpshrodinger Waiting for the fb!

[Aditya Jhalani @adityajhalani43]

@SahilPanhotra $20.40 is the wildest number here. at that cost, restricting weights wouldn’t keep the capability scarce for long.

[Aditya Jhalani @adityajhalani43]

@Harshalxh Waiting for the fb!

[Aditya Jhalani @adityajhalani43]

@M_jawad_yasin Waiting for the fb!

[Aditya Jhalani @adityajhalani43]

@disha_rathod24 working on music AI rn, mostly training models on jazz solos and judging by ear

[Aditya Jhalani @adityajhalani43]

@HashgraphOnline search and send email should not share the same permission.

[Aditya Jhalani @adityajhalani43]

@TheHashton building with ai and shipping side projects rn, so this is pretty much my corner of x

[Aditya Jhalani @adityajhalani43]

@mcuban @grok they might not even need a new attack. combining old weaknesses in a way nobody tested could be enough.

[Aditya Jhalani @adityajhalani43]

@PhotonCap the bigger bet here is owning more of the stack, not just selling the compute

[Aditya Jhalani @adityajhalani43]

@arena @OpenAI the cost gap is way more interesting than the rank here. being 0.09 points behind gpt-5.5 at 91% lower cost is kind of absurd