If he was honest with himself, he would rightly say he did those things for his own edification. If you have the time and ability. As someone who also cares deeply about his craft and spends more effort than I probably should improving my personal tooling, I'm clearly all for it.
Is it strictly necessary? Most certainly not. I've met equally dedicated and talented individuals who are fine adapting to existing tooling and working around idiosyncrasies rather than improving the tooling. That personally drives me batshit, but I can stomach it if required.
Maybe, but I can say without a doubt my reaction would be the same regardless of who the CEO is or the company. I think the only thing that matters for a CEO is business acumen, not moral character (and I'm personally moral af, often to a fault it feels like).
Notably real business acumen precludes lawbreaking, but until there is concrete evidence, I won't simply rely on the court of public opinion.
An important factor is that this man could not have sex without endangering his company. There were billions of women he could cheat on his wife with, and about 1000 who would be dangerous to his company to cheat on his wife with. He continually cheated in that pool of 1000.
Otherwise I don't care if this old man has sex, I don't even like thinking about it. But he was reckless and up his own ass about it. And while I don't think that the chips are the gold they're made out to be, and completely don't buy the reasoning of how A. Huang's business would be in the service of Chinese "spying," she's certainly someone who could and would talk business with him. She may have been an advisor on the financial house of cards for all we know, and the Chinese would definitely be interested in destroying the US economy.
Hell, the covert Chinese action might have been to take a woman who is extremely suspicious, get her to sleep with Huang, publicly out the affair, and let the company fall into the hands of somebody who is going to do something very stupid with it. That would still be Huang's fault. Maybe, none of this has anything to do with the Chinese at all, and the Israelis financed A. Huang's company (which was also evading Chinese enforcement) in return for her seducing a known sexually incontinent J. Huang, they publicized it, and with J. Huang's ouster the firm will fall under Israeli control. That would also be Huang's fault.
None of this excessively involves lawbreaking or morality, just stupidity and incompetence. Which is not "business acumen."
I blame my iPhone for a lot of it. In fact I purchased the Duo because I thought the bigger keyboard would reduce the prevalence of typos. It has, but still not quite enough. Apple really has screwed up the keyboard mightily in the past decade.
> Using privacy-preserving methods, we study 207,000 Claude and Claude Code research sessions to understand how Claude is used across scientific fields and research tasks.
Did they get permission to use researchers' data or did they monitor and extract data without consent? Considering the kerfuffle OpenAI generated with their math proofs, you'd think Anthropic would take a hint and be clear that they did the research with consent, though that assumes they got consent.
Seriously, all the industry researchers at these companies went through IRB training during their PhD; have they just conveniently forgotten?
I called out their experience and knowledge of Institutional Review Boards which govern human subjects research at universities. Embedding consent in a privacy policy would not count as informed consent [1].
Of course, only institutions receiving federal funding are subject to these rules, so companies like OpenAI and Anthropic are legally in the clear. I'm just saying, you'd think they'd have more awareness considering they've all been trained about informed consent during PhD.
Seatbelts shouldn't be mandated. People should have the choice, you just have to make the case that it's worth it. In fact some places don't mandate seatbelts (New Hampshire doesn't, aligning with their motto "Live Free or Die").
That said, for those who cannot advocate for themselves, like children, I think it's entirely reasonable to hold someone accountable if they decide to forgo an appropriate safety mechanism that results in harm.
I think there's a spectrum of enforcement. If they're mandated and you go to jail for not wearing one, that's stupid. If they're mandated and you get fined $20 if caught, just to add a little friction, is that still stupid?
I bet cars these days send a flag to the mothership when you drive without one, but the mothership doesn't pay attention to that signal unless it's linked to your insurance.
They haven't shipped yet (people who purchased the moment it became available have delivery dates near the end of the month). That said you can estimate it's speed by looking at the 256GB version for models with similar numbers of active parameters (GLM 5.3 has 40B active parameters). So for 4-bit you could compare to what people are getting Qwen3.8 27B at 8-bit (which is around 2000 tok/s prefill and 70-80tok/s token generation [1]). It might be a little higher or a little lower due to the different architecture and kernel optimizations.
If you don't like these experiments, then I can only assume you would complain about CRISPR as well, considering it's a general framework with many disparate parts that work together synergistically, similar to the work Takehi et al. describe. CRISPR was also famously rejected multiple times by reviewers.
Note that being able to both inspect raw weights and run Python code to perform additional analysis means the model has multiple possible techniques at its disposal (generally considered a positive, rather than a "confusing mix").
I don't understand CRISPR well enough to comment on the analogy.
The blog post claims to show that frontier LLMs can understand and craft neural network weights natively. The experiments where the model completes coding tasks are not relevant to this conclusion. Those experiments require the model to demonstrate an analytic understanding of how a network might behave under various conditions, along with reasoning skills, coding skills etc. As a coding task this is a significant challenge. But it's entirely different from "natively reading and writing neural network weights".
Would you also consider it a benefit if there were experiments showing that a frontier model can download and run PyTorch, or write its own network and optimiser? That's another technique it has at its disposal to improve an ML model. It's just not one that's relevant to the paper's core claims.
The blog post exhibits the same flawed logic that many academic reviews do — looking for reasons to reject a work rather than accept it. It comes off cynical and contrarian. It's easy to find issues with an academic paper, but ACL has consistently pushed for a higher standard [1].
Being the author of spaCy, he should know better. I certainly enjoyed the straightforward simplicity of spaCy when I used it in my research (though I haven't touched it in years).
You've expressed your criticism plainly without addressing anything the post might've done well. I think that's absolutely fine, but why would your criticism apply to the post but not your own comment? Personally I think it's because the standards of peer review don't apply to HN comments, but I don't think they apply to blog posts either.
Considering I have a PhD in CS focused in NLP and have been a Senior Area Chair for ACL, yes I do know how academia works. And, no this isn't science working; it's the same plague infecting most of academic blogging and social media. People are vying for attention in whatever way possible, often at the expense of others.
And the examples in the OP's blog post just go to show that reading comprehension is at an all time low:
> Just by looking at the raw weights, GPT-6 Astra can tell what a network was trained to do. And without any environment feedback, it has no trouble writing, from scratch, small game-playing networks that outplay networks trained with PPO for 100 million steps.
The table just below shows this statement to be *entirely* factual.
> Likewise, asked to make Qwen3.5-0.8B better at math without gradients, Astra identifies an issue in its weights causing reasoning loops that Qwen’s own model card warns about, and fixes this by natively editing 0.016% of the weights. With the same sampling settings, this weight correction raises MATH-500 from 7.9% to 48.1% and generalizes out of the box to instruction following, raising IFEval by 3x.
Any further implicit assumptions are on the reader, not the writer, as they never stated "that the Qwen experiment worked ‘just by looking at the raw weights’". You can certainly point out issues in the experimental design, but just stating you dislike the set of experiments and is yet another poor take that directly contradicts ACL's guidance for reviewing, exactly because it does not advance science.
All I can say is I very much disagree, and if I had authored the post and it had received the reaction from people that it has I would be anxious to issue corrections. I haven't reviewed for conferences for several years now, but I wouldn't have thought this should pass review.
I believe the post definitely plays off the implication that the Qwen experiments are connected to the experiments that read the raw weights and have no Python harness, when this isn't the case. There are experiments where Astra reads weights directly for very simple models, and there are experiments where Astra performs ML research tasks by analysing the problem and writing Python code. The authors imply that the same technique underlies these but that isn't the case.
Google has had exactly two quarters of negative free cash flow in its entire history as a public company, both of which occurred back-to-back in 2026. That's spooked investors who are already wondering if it's a bubble, which is leading to sell-offs across tech.
This isn't an everyday occurrence, so billing it as a panic every time there's a correction seems a bit off.
Weird. FastMail is many times faster than Proton ever was for me. FastMail is also headquartered in Australia, not the US. Fastmail also now has European servers.
Data regions like these offer no protection, it’s just a feel-good marketing term with no legal basis. Source: I work in this business with multinationals with a similar footprint and we have paid actual lawyers to look at it.
The only thing that unfortunately matters here is that Fastmail has a US incorporation.
Fastmail even writes as such:
> If what you need is a guarantee that your data remains only in the EU, we don’t have that, and we’d rather tell you directly than let you assume otherwise.
Is it strictly necessary? Most certainly not. I've met equally dedicated and talented individuals who are fine adapting to existing tooling and working around idiosyncrasies rather than improving the tooling. That personally drives me batshit, but I can stomach it if required.
reply