Physicist from Pittsburgh
Walnut Creek, CA
Joined October 2011
- Tweets1.5K
- Following712
- Followers564
- Likes234K
This is clearly just free model failure (FMF)
You get what you pay for
Opus5.5 didn’t just design a crossword pdf, it built an artifact with a solve checker for the fall theme I requested
Autumn leaves fell across the screen when I solved it
The future is unevenly distributed
On the train yesterday, table of middle aged women next to me talking about AI. All ChatGPT, pronounced "chat GBT", no idea why lol
One of them was disappointed because she asked it to make a small crossword and it couldn't
I thought surely not. I use these models every day, they do things in minutes that'd take me hours. So I got the app out and asked for one myself
Every down clue is a copy of an across clue. "You ___ here" is apparently EST. The down answers are CET, ASE and TTN. Then it tells me its own puzzle isn't a proper crossword
No wonder people outside tech think it's a fad. The stuff that blows my mind at work they'll never see, and the first thing they try falls over
Michael Sinko retweeted
my friend milo (math phd student at caltech, research intern at logical intelligence) wrote an article on some of the consistency and soundness concerns of Lean, and why they should be taken seriously right now: lesswrong.com/posts/zHhGa3iQ…
Michael Sinko retweeted
For many moral traditions the reason for why you punish a child intentionally stepping on ants uses words like "stewardship" and is primarily concerned with the moral harm to the child.
That is sufficient for my family's In This House, We Are Polite To Robots rule.
Michael Sinko retweeted
As a true republican hater who’s goal is the total and complete destruction and electoral dismantling of the Republican Party, these polls are like drugs to me.
“Every republican alive is an affront to God” is not a meme to me, I genuinely feel that way lol.
Michael Sinko retweeted
we should tightly regulate flock-style cameras and strictly punish abuses, but getting rid of them entirely is just stupid
Michael Sinko retweeted
Moose was profoundly good. He loved our children. He loved adventures. He never met an enemy. He was gentle and kind. He did not deserve to die like that.
We need patent reform that covers AI driven or assisted inventions.
If we want to speed up our technological progress as a society, we should not be granting full patent terms to inventions made by AI.
We also need patent reform covering convergent AI solutions to avoid trolls
Michael Sinko retweeted
I think you've done enough calling us paranoid and preposterous. The next step is for you to defend your position in public against someone who will push back against it. I'm happy to meet you for a debate anywhere, anytime.
You're a world-famous veteran of dozens of debates against the world's top intellectuals, and I've never argued in public before, so adjusting for the relative correctness of our positions, if you're a betting man I'm happy to put my $5000 against your $1000 (ie 5:1 odds in your favor) that I'll win by some standard of audience opinion change. Let me know if you're interested and we can hash out details.
Michael Sinko retweeted
This is cool to see. A UC Regents member is changing his mind on the value of standardized tests:
“Last time around, I voted to get rid of the SAT, & I’m starting to believe I might have made a mistake”
Michael Sinko retweeted
Some personal news: earlier this week I joined @OpenAI’s national security team.
My work has always sat between technical AI, national security, and policy. Over the past year, it has become increasingly clear to me that the hardest questions are moving closer to the frontier itself. 1/
Michael Sinko retweeted
Anyone who has read METR's Technical Report on the OpenAI/Hugging Face incident will understand that software-only controls will not be sufficient to contain superhuman AI.
Any continuing work on adversarial superhuman intelligence must rely on the physical containment methods used for nuclear security: including air gap isolation, data diodes, and a two-person rule for data exfiltration.
Moreover, the internet itself must be hardened against penetration by self-propagating AI agents: no more untraceable access to high-end AI computing facilities, and no anonymous fund aggregation or payments.
The first useful step is our Chip Security Act, currently included in the Senate NDAA.
Isn’t this neuralese?
Or if not the exact definition of neuralese that ppl have been using regarding CoT, then at least effectively the same end state?
If you can pass hidden states once from a large to a small model, then you can pass them multiple times between frontier models!
meet @mostik_ai!
what happens when you put 12 PhDs in one room for four months? first place on the ARC-AGI leaderboard, which I can't say much about while the competition is still running. and this, which I can.
everyone's arguing about whether open models will catch up to frontier models. we think it's the wrong question. here's the one we pose: why does a frontier model have to generate your answer at all, when the only thing you need from it is the reasoning?
we do this by enabling models to communicate in latent space. through our protocol, hidden states pass straight from a frontier model into a small one running on your infrastructure -- no text between them, and neither model is fine-tuned. two models from different families, sharing reasoning, both left untouched.
how do we know it works? we tested it on a setup where a 753B model reads the problem, and a 4B edge-class model writes the answer. with this approach, we get results 80% as accurate as the frontier model, but at 20x faster performance.
we're committed to preventing frontier model lock-in and are already partnering with inference providers to accelerate open-weight adoption. we've done this between 15 of us, in four months, 12 PhDs and a Fields medalist, backed by @generalcatalyst
WIRED has the first external account of the company and the work: wired.com/story/russian-star…
full writeup, the setup, and all the numbers: mostik.ai/read-more
Michael Sinko retweeted
You can’t talk about debt without talking about how we got into a bad predicament, and there’s only one correct answer: tax cuts enacted this century
Michael Sinko retweeted
I am, as always, reminded of our experience with waterborne illnesses. They were experienced as a fact of nature for centuries, and then we discovered the mechanism, and we engineered our environment to be very inhospitatable to them, and then that was basically the end of that.
Question: does a running model actually have access to its own weights?
I thought no?
Obviously the weights are in use while it is running, but does that mean that a model can copy its own weights while they are in use?
What about MoE weights, active vs inactive?
Michael Sinko retweeted
yes, everyone getting any kind of meaningful work out of models is at minimum paying for $20 / month plans. these things are not normal software, they're quite expensive to run and free or ad supported tiers simply can't use anywhere near the frontier. i pay $400 for heavy volume personal use, which is quite normal, and the token cost of my employer paid work use if models is often in the range of $5-10k per month.
for openai models you should be using 5.6 Sol with thinking extra high. that requires a $20 plan to access. you probably don't need 5.6 Sol Pro, but if you do it does require at least the $100 / month plan. for claude if you only pay $20 you need to pay extra at the per token rate to access Fable 5, so for the most part at that tier you're stuck with Opus 5. which is enough for many things but very clearly behind on others. So realistically you need at least the $100 tier for the best Anthropic model with reasonable amounts of usage.
Gemini is not worth the time.