MrOrelliOReilly

In this age of constant change, I am grateful that at least I can still rely on Google for confusing, ever-changing product names.

Chicanery! As others have stated, this is total AI slop. How do these things get upvoted?

This is a deflationary comment. You make a good point that the weights today are detectable, but I think proofs of concept around poison pills are very important. I am unaware of any theoretical reason why an attacker or misaligned model could not one day find a way to disguise a poison pill such that it is indistinguishable from the normal weights. There could be such limits, but research in this direction is very useful!

Hopefully it's benchmaxxing. I am terrified of the idea of Zuck leading the frontier of intelligence.

Does this response properly anticipate how math will change further with the next N model generations? Exposition and exploration may fall well within the capabilities of future models.

Tell me you you don’t use frontier models regularly without telling me you don’t use frontier models

The Last Witnesses by Svetlana Alexievich. A sad book!

I'm not sure I follow your logic. Paying for a service does not mean you get access to all potential services a provider offers. Providers can choose to keep some services internal.

Silly example: I pay Netflix for their most basic plan, so I get ads. Just because I already pay them money, doesn't mean I have a right to no ads! It also doesn't mean I have a right to 8k streaming; maybe Netflix reserves that for their internal cinema.

My read is that this is motivated purely by cybersecurity concerns. I don't have the impression that the whitehouse is suddenly x-risk pilled. Still, it's good to see the US taking steps towards regulation of powerful AI. Also a sign that regulation remains a topic with bipartisan interest.

I'm not really clear what it means to be designated a "covered frontier model" however? If it's a standard term, I haven't encountered it before.

Clearly AI-generated writing (confirmed with Pangram). Amazed this has so many upvotes—are people even reading the article?

@dang I know you have so far resisted a rule for AI-generated content (as we now have for comments), but I personally would prefer a flag for articles so that I don't waste my time on slop.

This article is pretty slim on details, but I agree with the general argument that dualism is unnecessary to explain phenomenal consciousness. The word "consciousness" has a lot of baggage, which causes us to mislabel cognition as consciousness. [1] This is why I really like using terms like "qualia" or _phenomenal_ consciousness to make explicit what we're talking about.

I still don't like this new trend of dismissing the hard problem altogether. We really don't have an explanation of phenomenal consciousness—it might even require novel physics to explain! [2]

I'd also like to point out that, though this might seem like a semantic argument, it has meaningful consequences for how we approach science and ethics. [3] For example, if we are physicalists and accept that phenomenal consciousness is a property of the world, what does this tell us about other unobservable properties of the world science may be missing? (Recall that we only know about phenomenal consciousness through our own experience of it; we cannot observe it in others)

[1] https://write.ianwsperber.com/p/what-is-the-color-blue

[2] https://youtu.be/DI6Hu-DhQwE?si=RB3qkt6PZ62SVpx3&t=2493

[3] https://write.ianwsperber.com/p/morality-without-consciousne...

> It’s just the caffeine

Fair enough if the use of “dopamine” is imprecise, but excessive screen time / doomscrolling / shitposting is definitely enough to wire you awake on its own, without caffeine.

Exactly!

Relevant Hacker News thread https://news.ycombinator.com/item?id=47904224

Ugh wrote “latter” and meant “former.” I didn’t mean lack of eng talent, but product

IMO this is the consequence of a relentless focus on feature development over core product refinement. I often have the impression that Anthropic would benefit from a few senior product people. Someone needs to lend them a copy of “Escaping the Build Trap.” Just because we _can_ rapidly add features now doesn’t mean we should.

PS I’m not referencing a well-known book to suggest the solution is trite product group think, but good product thinking is a talent separate from good engineering, and Anthropic seems short on the later recently

Yeah, good comparison. To Lerner's credit, he is always a poet at heart, which leads to concise, lyrical prose. DFW is voluminous in comparison; when it lands, its great, but it can feel overinflated/overdone when it doesn't.

I am a huge fan of Ben Lerner and have a copy of “Transcription” at home, waiting to be read. Autofiction is in many ways _the_ dominant mode of contemporary American literature, particularly among the literati of NYC/London (cf. Ocean Vuong, Tao Lin, Patricia Lockwood, etc., etc.). It can, for this reason, feel overdone and out of touch. But Lerner comes to the topic with such skill and intelligence, he really defines the genre for me, in a positive light.

> People hate AI so much that they are prone to attribute to it everything that’s going wrong in their lives, regardless of the truth. That’s why they mix real arguments, like data theft, with fake ones, like the water stuff. Employers do it, too. Most layoffs are not caused by AI, but it’s the perfect excuse to do something that’s otherwise socially reprehensible.

Pertinent quote. A lot of AI discourse goes in circles trying to evaluate the truthiness of every individual complaint about AI. Obviously it's good to ensure claims are factual! But I believe it misses a broader point that people are resistant to AI, often out of fear, and are grasping for strategies to exert control. Or at least that's my read of it.

Refuting individual claims won't make a difference if the underlying anxieties aren't addressed (e.g., if I lose my job will I be compensated, will we protect ourselves against x-risk, etc).

That's a fair response. But I'm not aware of any metrics supporting the point that the lag time is decreasing. The discourse I've seen has more focused on the ways Claude/OpenAI/Google have pulled away from the rest of the pack.

To be clear, I accept you might be right, but I think the crux is whether lag time is increasing, steady or growing.

How so? Opus and Sonnet are frontier models which cannot easily be replicated. Compute has real physical constraints which require appropriate procurement at this scale. At least those two points seem like pretty strong moats against the majority of companies.

I like the principle, but I also find that we software folk commonly mistake the creation of a website as the goal, rather than the production of "content" (e.g. blog posts). I spent years trying to publish a blog and continually getting derailed building the ultimate static website. Recently I switched to a Substack hosted on my own subdomain, and now I'm finally writing. At least I still own the subdomain.

I’m not convinced you read the post. I believe the author makes quite explicit their goal was to actually visit these cities, noting this is far from the most efficient bus route. Their itinerary also shows long stays in several spots.

I have had the same hypothesis around the recent operational success of US military interventions, but would agree with other comments here that this is more "vibes" than data. It's been reported that Maven (integrated with Claude) has been used extensively for Iran, but I haven't seen any hard evidence this is directly contributing to greater US military efficiency. I do buy the general thesis that AI would support operation excellence and solve attention problems across concurrent actions. Would be good to see some more reporting or combat analysis to try to measure the contributions of AI (e.g., how many more concurrent aerial sorties are taking place vs equivalent interventions, how many more strikes are "successful" vs past, etc).

EDIT: I see this post has been flagged. Why? I understand it’s political but it seems very much within the site’s ethos. I didn’t get the impression it was AI-writing either.

Correct, I am stating that the stochastic parrot hypothesis is a fallacy.

I think you might be underrating the value of even that enabling work. Some parents would not have the financial resources to provide those learning materials. And some parents would take a normative stance on how an 8 year old ought to behave.

More importantly, it's not as though individuals like Clements or Erdos was corresponding with Terrence directly to arrange a meeting. His parents clearly played an important role in facilitating and allowing these encounters. That deserves a lot of credit!

This all sounds like the stochastic parrot fallacy. Total determinism is not the goal, and it not a prerequisite for general intelligence. As you allude to above, humans are also not fully deterministic. I don't see what hard theoretical barriers you've presented toward AGI or future ASI.

I don’t believe the article makes any claims on the infeasibility of a future ASI. It just explores likely failure modes.

It is fine to be worried about both alignment risks and economic inequality. The world is complex, there are many problems all at once, we don’t have to promote one at the cost of the other.

Totally agreed. Most the weird concepts of Gas Town are just workarounds for bad behavior in Claude or the underlying models. Anthropic is in the best position to get their own model to adhere to orchestration steps, obviating the need for these extra layers. Beyond that, there shouldn’t actually be much to orchestration beyond a solid messaging and task management implementation.

The author's high-value flowcharts vs Steve Yegge's AI art is enough of a case-in-point for how confusing his posts and repos are. However this is a pervasive problem with AI coding tools. Unsurprisingly, the creators of these tools are also the most bullish about agentic coding, so the source code shows the consequences. Even Claude Code itself seems to experience an unusually high number of regressions or undocumented changes for such a widely used product. I had the same problem when recently trying to understand the details of spec-kit or sprites from their docs. Still, I agree that Gas Town is a very instructive example of what the future of AI coding will look like. I'm confident mature orchestration workflows will arrive in 2026.