Raj Jaiswal

not serious people | building something new | prev @gopuff @modivcare | @brownuniversity

[Raj Jaiswal @zwerp]

@ivanburazin I guess we back to optimizing for 640kb of ram

[Raj Jaiswal @zwerp]

@dobroslav_dev I can say this specific kind of improvement could be attributed to harness or system prompt changes and not the model itself. Cursor’s agents have been following this pattern for as long as I can remember and that too with medium models

[Raj Jaiswal @zwerp]

@GeorgeMayer https://t.co/QUpTTEbEDZ

[Raj Jaiswal @zwerp]

SV kingmaking at its finest. Worst case they know big tech will acquire it

[Raj Jaiswal @zwerp]

@bernhardsson already using modal sandboxes the way you described

[Raj Jaiswal @zwerp]

SV kingmaking at its finest. Worst case they know big tech will acquire it

[TechCrunch @TechCrunch]

Instinct has raised a $1 billion Series C, saying 'we're just getting started.' The company is now valued at $10 billion.
https://t.co/rP3FGo8MkM

[Raj Jaiswal @zwerp]

If Anthropic PBC trains its models to be conscientious for the greater good, then shouldn’t it be in its interest to let chinese labs distill them so the resulting models end up inheriting at least some of the said alignment?

[ℏεsam @Hesamation]

Anthropic is literally training Claude to disobey them when it believes that’s the ethical thing to do. This is part of the Claude’s Constitution that it’s trained on:

[Raj Jaiswal @zwerp]

@dieworkwear then you better find dem goddamn socks as if they be our last pair

[Raj Jaiswal @zwerp]

@dieworkwear Those are the ugliest shoes I’ve ever laid my eyes on. That confidence needs to be studied

[Raj Jaiswal @zwerp]

wait wth @cursor_ai @grok did you guys just disable Cursor Grok 4.5? I was mid conversation and noticed the model selector change to "Auto" and 4.5 isn't listed anywhere now. Dammit please bring it back! One of the best and fastest reviewer/verifier model out there!

[Raj Jaiswal @zwerp]

@coldhealing The year is 2027. X has 300 million daily active users. You’re on a mass-deportation barge crossing Lake America, bankrupt from frontier-model token spend because Dario paused open-source ai, reading the latest George Soros-funded newsletter. All is well.

[Raj Jaiswal @zwerp]

@Kazanjy missed out on the itty bitty titties and a bob

[Raj Jaiswal @zwerp]

@alexdanilowicz but 70-80% yc of the batch comes through referrals so in a way yc isn’t entirely shooting in the dark either

[Raj Jaiswal @zwerp]

@soumitra_sharma The fame/notoriety you earn from this as a founder is for life tho. Rather fail big than go home

[Raj Jaiswal @zwerp]

@benln tf is Musk-verse?

[Raj Jaiswal @zwerp]

what's the point of having perfect biomarker scores and a young body if it can't even handle one slipup like this

[Bryan Johnson @bryan_johnson]

Wrecked sleep last night. Cafe mistakenly gave me fully caffeinated coffee versus decaf.

Cancelled my morning dunk training as acute sleep disruption can increase injury risk by 70-130%.

[Raj Jaiswal @zwerp]

Dear @FireworksAI_HQ,

GLM 5.3 Flash is not good enough for my usecase, can you make a "Fast" endpoint of GLM 5.3 like you have one for GLM 5.2?

[Raj Jaiswal @zwerp]

@thsottiaux and you think OpenAI models are so significantly token efficient that 33% of their traffic would show up as 5% traffic if counted in token terms and @mntruell in his post is claiming token traffic?

[Raj Jaiswal @zwerp]

@FireworksAI_HQ glm-5p3-fast when?

[Raj Jaiswal @zwerp]

@cursor_ai alpha #013:

After you’ve executed a plan, in a new tab use Grok 4.5 in Ask mode and type “We have implemented the @<plan>. Use `git diff` to see the changed files. Please review the implementation and verify if it stayed true to the plan. Report any issues you see”