Shawn Thuris retweeted
So I recently posted on here that one of the reasons I'm opposed to an AI pause is that I don't want to die.
An entire thundering horde of blithering idiots immediately started screaming at me that I'm a selfish bastard who wants to burn the world down.
These people seem to assume I have some kind of supervillain bunker I plan to retreat into, cackling madly and clutching my extended lifespan to my chest, rather than having to live in the same future world as everybody else.
This is a deranged response. I had to think for a while to even begin to mentally model the kind of bugfuck craziness that could produce such thinking.
Now that I think I have done so, I have a public service announcement to make:
Some people will try to convince you that life extension, or wealth in general, is a zero-sum game. Then in order for some to win, others must lose. For any individual person to achieve an extremely long lifetime, the world must be plunged into misery for others.
These people are not your friends. They are not saying these things to help you or save you. They are telling lies which, if you believe them, will control you. And control is their goal.
Did we learn nothing from the last century? The self-anointed scream that a vast catastrophe is heading towards us all, unless we surrender our wealth and liberty to them. The label on the catastrophe changes: overpopulation, pollution, global warming, rogue AIs. The prescription never changes. It's always a grift, always a lie, always a power-grab.
Don't get fooled again.
Shawn Thuris retweeted
In my ongoing quest to find the limits of Opus 5.5, I directed it to audit my old scientific field, theoretical astrophysics. We indexed about 150,000 papers, audited 50,000 for issues, and did progressively more thorough checks on thousands more. We found nearly 1000 major errors that got cited many times, most are still uncorrected. This survey enabled me to estimate the efficacy of peer review for correcting errors (not very good) and to identify dozens of open questions that anyone with spare Claude Max tokens could answer today (.md context files included).
Bottom line: AI can speed up quantitative science by >1000x with no further model improvement required. But I'm seeing no evidence of this being pushed within universities, only academics complaining about labs dropping math proofs.
Among my fellow reformed academic high agency founders, however, several are speed running entire disciplines.
This survey has implications for how science will be run in the very near future. Hint: Academic publishing is dead. It was already a very stupid way to record and promulgate knowledge, and had been so for decades, even before all that publicly funded knowledge was locked behind paywall cartels. Now AIs can generate thousands of novel, correct papers per day, it is no longer meaningful for humans to consider a research paper a "unit of science" or to even bother much with the 99.9% of science which, up to this point, was necessary drudgery. Time to discover something like General Relativity or Evolution every week. LFG!!
Enjoy: science.caseyhandmer.com/phy…
Heading for the catch attempt in the next 15 minutes
Watch Starship Flight 15
teapawt.net/i/broadcasts/2brajgZlO…
15
Shawn Thuris retweeted
A large sneaker wave crashes over a seawall in Pacifica on Saturday. Buoys offshore measured the largest waves in more than a decade, topping 30 feet.
Read more >> sfchronicle.com/weather-fore…
Shawn Thuris retweeted
Got more details, it was NOT DNS, 100%. Also doesn't look like AI for those asking. And NOT "layoffs related" lol. But a seemingly complex set of networking problems caused by the legacy system
Mitigation was v complex: not just rollback network config changes, but eg registry key patches, rebooting specific racks, and dealing with the thundering herd problem via traffic routing
the root cause is still unclear even internally within MSFT (save for the incident team) and everyone awaits the postmortem eagerly. This will surely be a public one. I'll see if I can get details on top of the public postmortem from MSFT
Shawn Thuris retweeted
Introducing e-means, an algorithm beating k-means in 10x fewer epochs.
With order of magnitude wall-clock speedups, e-means is released in FAISS and needs no hyperparameter tuning beyond k. It is robust to various scales and datasets. Still skeptical ? Just try e-means [1/10]🧵
These points accord with my limited use of Fable 5.2 so far:
'- it patronises more than Opus does and is leaning towards Astra in semantic BS. - Anything knowledge dependent it’s a monster in. Problem with Claude models is they hate researching... And even when they do research, they enforce their own beliefs in the research... So having a model that knows more leads to better results especially in planning. - It is SOOO good to speak with and debate... Original Fable 5 and 5.2 are the only two instances I’ve ever had where it raised a point and I said “holy shit, it’s right" '
48 hours opinions on Fable 5.2:
Bad:
- BURNS limits. Way worse than Fable 5.1. Xhigh burns faster than Max did on 5.1.
- It’s slow unless you use fast mode and then it’s normal but ridiculously expensive. Burned $500 in credits in 20 minutes.
- It makes really poor decisions and infers intent worse than Opus.
- it patronises more than Opus does and is leaning towards Astra in semantic BS.
- It has really low lows. It’s worst 10% is way worse than Opus.
Good:
- At its best it’s a lot better than Opus. It feels as if it has cleaner follow through and strings things together better.
- It’s a different breed at quantitative and data analysis. Opus was just stupid at this. Especially if it required market or competition understanding or even a psychological understanding.
- Anything knowledge dependant it’s a monster in. Problem with Claud models is they hate researching, so inherent knowledge is especially important. And even when they do research, they enforce their own beliefs in the research their search queries are targeted and just wrong. So having a model that knows more leads to better results especially in planning.
- It is SOOO good to speak with and debate. The lowest sycophancy rate I’ve ever seen and it has the typical “Fable” smell of great knowledgable and intelligent arguments. Original Fable 5 and 5.2 are the only two instances I’ve ever had where it raised a point and I said “holy shit, it’s right”. First example was an interpersonal debate and second one was market dynamics. Every other model I scoff at because the points it raises can always be easily refuted or missed important factors.
- It is a much better assistant than Opus. It goes “above and beyond” when Opus would never. Like setting reminders for itself or checking the status of an App Store release and messaging me when it’s ready in Hermes.
1
1
72
They post-trained one 27B model not to take the CCP line on certain sensitive topics and people are freaking out. Just use a different model! There are 6,672 other models of that size to choose from on HF.
23
Nice DeepSWE score! Does it survive a lowercase filename?
Introducing MaxxBench: 89 DeepSWE tasks with small changes to the instructions or starting code. Same engineering problem, slightly different requirement.
Models clustered around 74% on DeepSWE end up 23 pp apart.
Good at coding, or good at DeepSWE? 🧵
Synthetic DNA has incredible potential but it's also a potential tool for bioterrorism. This policy will help to keep it limited to constructive ends.
Today, we released the National Policy for Nucleic Acid Synthesis Screening to protect American biotechnology innovation while reducing the risk of misuse.
This policy establishes a new, government-wide screening requirements for synthetic DNA and the equipment used to make it. It will ensure companies selling nucleic acid sequences and synthesis equipment know their customers and prevent the malicious use of these materials and tools.
This policy strengthens our Nation’s biosecurity posture by establishing rigorous U.S. government-wide procurement screening and physical security standards for synthetic biology. It replaces outdated “self-attestation” systems with robust, independent verification. It creates a dynamic, SI-powered screening layer that assesses the biological risk from nucleic acid sequences, allowing the U.S. to anticipate and prevent malicious use and help stop dangerous gain-of-function research.
@NIST will issue rigorous new screening standards and @HHSGov and @USDA will certify synthetic biology providers. Together, these measures will strengthen American biosecurity, safeguard responsible innovation, and establish a global standard for safe and transparent synthetic biology tools and research.
1
54
Shawn Thuris retweeted
The irony of killing AI data centers to feed a bloated healthcare system instead is that you are actually killing people relative to the world where you let AI progress unimpeded. – at Port St John, FL



























