the worst thing you can do is not live your life because someone said that someday something might happen
Joined December 2025
- Tweets542
- Following524
- Followers105
- Likes1.5K
Faan Rossouw retweeted
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com/
Faan Rossouw retweeted
Daniel Martin Diaz, Biomimicry. Date unknown.
Graphite, watercolors and inks on archival paper.
Faan Rossouw retweeted
a=(y,d=mag(k=(4+sin(y))*cos(i/7),e=y/5-11)-6)=>point((q=99+3*sin(7/k)+y/23*k*(e+sin(e*9)))*sin(c=d/4-t/4+i%2*9+sin(t*3-e*2-d)/9)+200,q*cos(c)+200)
t=0,draw=$=>{t||createCanvas(w=400,w);background(9).stroke(w,66);for(t+=PI/120,i=2e4;i--;)a(i/598)}//#つぶやきProcessing
Faan Rossouw retweeted
Remember 2 days ago when half of X was ballwashing Zuck over his vaporware glasses. Gottem
The rule I keep coming back to when I build agent harnesses: only make things as complex as they need to be, starting with whether a job needs a model at all. Why codify something so elementary? Check the video.
The article, with the video: aionsec.ai/articles/the-gold…
It's from a lesson in my course, Build Your Own Agentic Defense: aionsec.ai/fundamentals
Faan Rossouw retweeted
AI lab: “our model escaped containment. this is terrifying.”
Everyone else: “holy shit, how?”
AI lab: “well technically the containment had a route to the internet and we gave the model hacking tasks.”
Fucking incredible.
OpenAI and Anthropic oversold AI security breaches to pressure feds into protecting turf: insiders trib.al/Jo6xMk5
Faan Rossouw retweeted
Build your own harness, folks.
This is absolute banger paper from NVIDIA on self-evolving agent harnesses.
(bookmark it)
They introduce SoL-Pi which cuts token traffic by nearly half.
And it matches its baseline harness on GPT-5.6 Sol and Opus 5.
More details below:
Instead of tuning a harness by hand, they run auto-research loops at the harness layer across many repository-derived and verifier-driven environments, keeping only the mechanisms that survive selection.
Four mechanisms survived:
> Action Fusion changes how actions execute
> Online Context Compact handles compaction during a run
> ObservationPack reshapes observation handling
> Evidence-Preserving Reducer covers delegated reading
On the 51-task EdgeBench evaluation, the savings translate to about a third off API cost. In dollars that is an estimated $8.75 to $13.50 per hour against native Codex and Claude Code harnesses, and $4.36 to $5.71 against the baseline harness.
Because the search runs across many environments rather than one, the retained mechanisms keep working outside the setting that produced them. Code is on GitHub under NVlabs.
Paper: arxiv.org/abs/2609.20519
Chat with Paper: academy.dair.ai/papers/sol-p…
Faan Rossouw retweeted
Anthropic has confidentially submitted a draft S-1 registration statement to the Securities and Exchange Commission.
Pending completion of SEC review, this gives us the option to pursue an initial public offering.
Read more: anthropic.com/news/confident…
Faan Rossouw retweeted
a=(x,y,d=mag(k=4*cos(x/21),e=y/8-20))=>circle((q=3*sin(k*2)+.3/k+sin(y/19)*k*(9+2*sin(e*14-d*3+t*2)))+50*cos(c=d-t)+200,q*sin(c)+d*39-475,k*k>15?2:1)
t=0,draw=$=>{t||createCanvas(w=400,w);background(9).noStroke().fill(w,116);for(t+=PI/240,i=1e4;i--;)a(i,i/235)}#つぶやきProcessing
