head of engineering @liquidtrading • i like music and building things
New York, NY
Joined June 2023
- Tweets2K
- Following275
- Followers375
- Likes10.1K
super cool seeing companies like @Cloudflare launching Zero Wire and @vercel launching Security Review so that you can be prepared for when Railway tries to hack you
the boundaries between software and hardware will only continue to blend, bullish on more personalized manufacturing solutions
eternal sunshine of the spotless model weights
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
on our way to intelligence too cheap to meter
Day 25/
We are doubling all usage limits for all Pro 100/200/500 tiers and Business Premium, effective today.
We heard your kind feedback and won’t be reducing usage limits for Pro 200, so your Work and Codex usage will stay the same. We’re also going to continue offering extended limits for dots for now, and usage of your dot will still continue to be separate from your default usage.
And we'll of course also keep shipping better models and other efficiency improvements that increase your effective usage over time.
one of the most rewarding parts of software development is identifying bugs in other people's code and getting to help fix them
i'm not saying its a bubble but...
with Instinct is reportedly raising at $30B some frame of reference seems justified:
- more than half of Delta Airlines ($53B)
- bigger than American Airlines, Southwest Airlines combined (~$29B)
- bigger than Snap + Pinterest + Lyft combined (~$28B)
- more than Reddit (~$26B)
i'm bullish on AI having positive effects on the economy in the long run but some of these numbers just aren't making sense
time to start smoking cigarettes guys
Over the past few months, Anthropic scientists have engaged in deep research partnerships with six organizations developing new medicines. We've also begun drug discovery efforts of our own.
In doing so, we’ve learned much more about how Claude can support the many stages of drug development.
In this blog post, we share some early results from these efforts, and describe how this work fits in with our wider ambitions in the life sciences.
anthropic.com/research/early…
i am so tired of the default link click behavior in software dev tools changing to opening them in some embedded browser
yeahhhh this does not bode well
“the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing”
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
Raymond retweeted
After the tsunami of security issues last week, I've been running security audits through my own codebases with Opus 5.5. By carefully prompting it to look for certain classes of issues in code that I own, and making that very clear, this seems to get past the guardrails. Claude found a few serious issues that I quickly fixed.
Everyone should be doing this now, we have the same weapons for offense and defense, so be sure you wield them first.
I tried to get Claude to confirm a security issue by testing it with my account. It would not do it, that tripped the guardrails no matter how I phrased it. It was willing to give me step-by-step instructions for how to test the exploit myself. Which was good enough.
