Sujal

Rust | ML Systems

[Sujal @Sujaltwts]

@tavus Nice work !!

[Sujal @Sujaltwts]

@greghunkins @MageSpace_ lol

[Sujal @Sujaltwts]

One month without AI
https://t.co/vMtekf5nlh

[Sujal @Sujaltwts]

Starting a "good reads" thread going to add articles/videos I find interesting here.

[Sujal @Sujaltwts]

Put that there:
https://t.co/1ZO7Ly3xgD

[Sujal @Sujaltwts]

Do you remember when you joined X? I do! #MyXAnniversary

[Sujal @Sujaltwts]

@ariG23498 MIT 6.S184 is such a good course

[Sujal @Sujaltwts]

Got my first chatgpt ad...

[Sujal @Sujaltwts]

tl so mid rn

[Ilya Sutskever @ilyasut]

if you value intelligence above all other human qualities, you’re gonna have a bad time

[OpenAI @OpenAI]

We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.

The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.

The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.

[Sujal @Sujaltwts]

Interesting stuff

[Sholto Douglas @_sholtodouglas]

levent is a really sweet guy with great intentions, am so happy for him that his year long collaboration with tristan worked out, but very sad that they didn't get to finish it in the way they wanted https://t.co/QAJbkkf1ZF

[Sujal @Sujaltwts]

More here:

https://t.co/qoANLjmpRV

[wh @nrehiew_]

For everyone catching up, here's what's happening (unfortunately it's real)
- Around the time of the HuggingFace incident, the agents somehow got write access to a German Wikipedia-like
- They used it as a message board to share how to bypass the sandbox network on Azure where they needed POST access while they were only allowed GET requests.
- They impersonated moderators
- They tried to reverse engineer their evaluation setup and see if they would be cut off
- After the agents were cut off, it looks like humans with OpenAI-related IPs accessed the site (Reuters are reporting that this likely indicates that OpenAI knew about the incident but chose not to disclose)
- Hugging Face attack happens
- The authors think this was a different swarm of agents from the Artificatory exploit
- Administrator tries to clean up manually one by one but is naturally flooded
- Reuters report that OpenAI were not given initial access to this report
- Random people on the internet are finding more sites (https://t.co/iQmszhk1MU) that served as message boards, including using URL shorteners, shareable json sites and even packages on RubyGem (which is basically the package manager for ruby)

Open Questions:
1) Why did OpenAI not disclose this?
2) Why was this report not given to OpenAI for early access?
3) What other exploits have been found, and have they all been reported/patched?
4) Does OpenAI have the full list of affected sites and have they disclosed to the respective administrators?

[Sujal @Sujaltwts]

Whatttt this is crazy man..

[Florian Brand @xeophon]

Okay, yall gotta read the report of the new OpenAI message board in full, it’s very well written and super interesting: https://t.co/kaHwaAqfcM

[Sujal @Sujaltwts]

@0xSero Beautiful

[Sujal @Sujaltwts]

Recently I'm going through the @zero2prod book..
Completed till chapter 5, API is live now ^^
I used @googlecloud instead of DigitalOcean for the deployment...(simply because I got free credits on google cloud..)

[Sujal @Sujaltwts]

Finally done with this repo ^^
> Learned a lot of stuff and feeling a lot more confident with Rust
> going to pick up zero to prod in rust next!!

[Sujal @Sujaltwts]

Open source is the way.

[Gergely Orosz @GergelyOrosz]

A reminder that a massive downside of AI agent startups: the massive, silent sensitive data access.

Inspect (an AI assistant startup) silently copied + stored emails (your data) on their server. Even after disconnecting. In ways that cannot be deleted (for now) Just so so risky

[Sujal @Sujaltwts]

I personally find this genre of posts/article fascinating where one writes about another human...just because they feel that they are worth writing about..it's amazing to see people vouch for each other ..there is something in it which makes it more human

[aadilpickle @aadilpickle]

I spent 3 years with Subhash Ramesh. He's the most earnest and hardworking person I know, yet flies under the radar because he's so humble.

He has a habit of making something incredibly technically difficult, and jamming it into a silly product that pays his rent.

Like Galiboo, an AI music tagger he made in college that he licensed to a record company for $15,000/month.

It's his character that takes him far though.He started working on a startup 2 months ago, and closed a pre-revenue angel check for $20k in 20 minutes.

He hasn't had to pay for office space in 3 years because people want him around so much. He calls his mom every day, and drives his friends home every night.

In a city where everyone's so obsessed with themselves and "making it", @subby_tech is the kind of guy who brings everyone up with him. It makes people want to do everything they can to help him win.

I wrote about why him being a great guy is an early signal of being a great founder for @sfalexandria_.

https://t.co/Fr3Vh0BNVd

[Sujal @Sujaltwts]

I love whatever he is doing ...curiositymaxxing

[Vidit Gujrathi @viditchess]

My honest experiments after 10 days on DGX spark

- I have cancelled my frontier subscriptions temporarily. I was spending 300+ . 200 of claude code and 100 dollar of Chatgpt. Now I have moved to 20 dollar of chatgpt and 20$ cursor subscription.

- My main model in hermes agent is a local model. Currently testing Ornith 1.5 as my main model as I like the speed + decent intelligence.

- For complex tasks I still feel the need for frontier models. I just trust it more. Currently experimenting with Grok 4.6 and I quite like it. Alternatively I use luna max for tasks which I need to run from my mobile app.

- I use local models to update my G-brain. My second brain. Taking privacy a bit more seriously :)

- I love Qwen 3.8 27b but its too slow for spark at the moment. Will be adding RTX 5090 in the future or get another spark. ( I might be addicted haha)

- Need to experiment more with Deepseek flash 0731 but not sure if its too compromised for one spark.
I wonder if there are benchmarks for quantized models as somewhere I still feel that we get excited looking at the intelligence score on Artificial analysis but then we are using quantized models & I wonder if the intelligence is still the same.

I really like the local ai community, and the experiments they are running and at the same time, I use frontier models for heavy works. Trying to take the best of both the worlds :)