AnodicElegy

I don't think the public is going to swallow subsidies for the richest companies on Earth so that they can avoid competition.

"...A proprietary model that the U.S. government is treating like something just shy of a nuclear weapon..."

This is more than a little hyperbolic. Maybe "just shy of Valium"?

"Prompts and completions are retained by the provider and are not used for training..."

I'm curious what the model provider is using the prompt/response pairs for, in that case. They aren't offering a model for free without their name on it for no reason.

Yes, they have multiple levels of Claude, GPT, Gemini, and Kimi, but not the other top models (I would put GLM, Qwen, Muse, Grok, and Deepseek in that bucket).

I understand that running these benchmarks can get expensive, but it would be really nice to see AA include more benchmarks of models at reasoning settings other than the maximum, at least for the biggest releases. They have that nice graph of cost vs. composite benchmark score with the Pareto frontier line, but who knows if those are actually the optimal choices? There are already a few non-max-reasoning models on the Pareto line, among the few that were tested.

"I wrote Machines of Loving Grace because I didn’t feel the AI industry was painting an inspiring enough picture of how the technology could radically transform the world for the better. The bulk of the essay is devoted to refuting skepticism of AI’s potential in health and biology, and showing why I think it will actually be possible to cure most human disease in ~5-10 years, as crazy as it may sound to ordinary people and frankly to biologists as well (I used to be one!). And, if you read my most recent essay (Policy on the AI Exponential), I discuss concrete proposals for how to streamline the FDA process to make sure the deluge of AI-accelerated drugs isn’t slowed down by the regulatory process."

This is fantasy, and I believe the vast majority of those actually working in pharmaceutical R&D would agree. Contrast Amodei's opinion here to the Derek Lowe post I shared recently: https://news.ycombinator.com/item?id=49313367 . Derek Lowe's opinion is closest to my own experience: AI, whether it be traditional ML or LLMs, can help here and there -- the former to help you to triage paths to explore in a way that's a little better than intuition in some cases, the latter mostly to generate code faster in the code-dependent aspects of pharma research -- but neither of these things are significantly widening the main bottlenecks. I don't think the data exists to do so, especially since so much of pharma R&D is looking for higher and higher hanging fruit (i.e. exploring avenues for which a trove of relevant training data does not already exist).

Interesting fellow. He has publicly claimed that he has ESP:

https://www.splcenter.org/resources/hatewatch/wikipedia-wars...

https://openpsych.net/forums/18/thread/25/?page=1#124

OP here: the title of the thread still links to Derek Lowe's blog post, but the article discussed in the blog post was added to the body of the original post (not by me).

Looks lovely. I wonder if it has good fishing.

Domnauer is overstating the "little people" effect here. Seeing miniaturized people seems to be only one of several manifestations associated with these mushrooms. This Wikipedia article (https://en.wikipedia.org/wiki/Hallucinogenic_bolete_mushroom) is a lot more informative. To me, it sounds like it could actually be a serotonergic substance, but it's not clear. I hope they find the active constituent!

I think a lot of people are simply glad to see so much viable competition in the LLM space. Given the oligopolistic outcomes in Big Tech over the past couple decades, it would be nice if Big AI turned out differently, even if a lot of those competing have monopolies or near-monopolies in other spheres.

That reminded me a lot more of Chip's Challenge than of mowing a lawn. Just missing some blocks to push!

That seems like a problem orders of magnitude harder than making a humanoid robot. We haven't even figured out how to make hamburgers without cows at a marketable price yet.

I imagine that the perception might be that a judge will approach the case more rationally and technically, whereas the jury might be swayed by sympathy for the victim, biases, etc. To what extent that's true I do not know, but conviction rates are significantly lower in Canada than in the U.S.

For all but minor offences in Canada, you have the right to a jury trial. Here, the accused elected to be tried by a judge. Most people do.

Let's not forget the good old Single Greatest Predictor ( https://www.philosophicaleconomics.com/2013/12/the-single-gr... ), which hit an all-time high in Q4 2025 ( https://fred.stlouisfed.org/graph/?g=1Wc2g ).

The safety ratios listed in reference [1] are exaggerated, as can be determined by the following qualification: "The majority of published reports of acute lethal toxicity indicate that the decedent used a co-intoxicant (most often alcohol)."

The vast majority of so-called drug overdoses are due to polydrug intoxication. It's much harder to die by consuming a single substance. This is clear from the statistics of the few jurisdictions that report all substances in the blood in coroner's reports. See, for example, the Scottish data from 2020 (https://www.drugsandalcohol.ie/34642/7/ndrdd_report.pdf ): "In 2020, almost all (96%) [drug-related deaths] occurred after the consumption of multiple substances."

If an LLM is used in the drafting of an article, this should, at least, be disclosed, preferably at the beginning of the article. For example, I recently came across this article ( https://thedispatch.com/article/affordability-crisis-healthc... ). The LLM voice was suppressed well enough until this inane passage:

"One number. Four completely different stories. The number is engineered to include all of them, because including all of them is what produces the 49 percent."

I decided to fact-check a statement ("CNN’s May 2026 survey found the share of Americans spontaneously naming gas prices as their top economic problem rose from 5 percent to 23 percent in a single year, with food costs cited almost as often") and it was incorrect (food costs were in fact cited more often than gas prices). Since the first thing I checked was wrong, I decided it wasn't worth my time reading the rest of the article. It was, as they say these days, slop.

It felt like a little bit of my time had been stolen. If a disclosure had been at the top, it would have been more of a caveat emptor situation.

It shocks me that we use to go to wave pools like these for elementary school trips. They never even asked us if we could swim or not. I was at least a passable swimmer, but even so, it was so crowded that I remember a couple times getting stuck under the water briefly trying to find my way up between all the feet and inner tubes.

Anthropic did the same thing recently:

https://www.anthropic.com/research/making-claude-a-chemist

I don't know why they think this is okay, any more than it would be to call their models "AI doctors" or "AI lawyers".

Ironically, where I live, most actual medicinal chemists are not allowed to call themselves chemists since you have to pay dues to a professional association to use that title.

Yeah, it's hard to see exactly what was gained here versus standard, machine-learning-based high-throughput experimentation.

This stuff drives me nuts. Chrome on desktop is another abhorrent example of this. Every time I log in to a Google account within the browser, it nags me to link the account to the browser itself. One time my partner clicked the popup that asks to link the account by accident (this is very easy to do and probably Google's intention) and it hijacked all of my bookmarks. When I logged out of the account, everything was gone. I looked into Firefox and Edge, but they have the same profile crap. I don't need or want to log in to my browser!

I've hit my limit with Microsoft, too. I'm going to start moving my PCs over to Linux. The last time I installed Linux on a personal device was in high school, almost two decades ago, and it took quite a bit of work to get everything running properly. I've heard that compatibility issues are much less prevalent now. If things to smoothly, I may even convert my main gaming and work PC, although I'll have to look more thoroughly into whether all my Steam games will be playable.

Out of curiosity, I pasted an article in French I was reading a few minutes before coming across this thread into ChatGPT and asked for a translation into English. It was certainly passable from a functional perspective, and I wouldn't hesitate to use it to translate an article from a language I don't understand. But it was not professional-quality work. There were a couple instances where the French grammar was mistranslated, and the writing was perfunctory, not going into any effort to have the article flow like it was originally written in English instead of simply translating each sentence literally. Would I read an article written like this? A short one. A novel? Definitely not.

I believe it's a huge advantage to be an avid reader, especially of novels, in childhood. It helps not just with the ability to understand and retain written information, but also with increasing eloquence and breadth of knowledge.

I read frequently to my two-year-old son. It's a fun challenge for me, as I attempt to give every character a different voice and keep it consistent, to emulate the professionals who narrate the audiobooks that I listen to. Currently, he is enjoying Archie comics. I hope the effort I'm putting into it now will help secure his attraction to the medium once he's reading independently.

I love video games as much as I do novels. My parents restricted them for me severely (along with all other screen time) as a child, though, so the amount of time I spent on them was much smaller than the amount I spent reading novels. I thank my folks for that now. I'm not going to be as strict as they were with my son, but I hope I can find a happy medium. He's already very attracted to television, but we try to keep that to short, occasional sessions.

I would hope that Fiverr is not a representative example of a typical employment process. I understand you have to show that you're not a laggard when it comes to communication, but being dismissed for not replying to a message in 5-10 minutes is insane.

"AI is advancing at a lightning pace—in only four years, AI models have gone from barely being able to write a coherent line of code to writing most of the code at major AI companies. Similar gains have been made in biology, physics, math, finance, law, translation, and many other fields."

This is a massive exaggeration. The advancement in the automation of computer code writing has been impressive and is obviously, at least in the short term, changing the software engineering industry substantially. Most other fields have not been affected to nearly the same degree. Certainly not biology, physics, finance, and law (I don't know enough about the math and translation fields to speak to those).

---

"3. Accelerating AI’s positive impact..."

This whole section is the type of thing that often comes out of the mouths of Silicon Valley tech executives without a pharma background. It indicates a thorough lack of understanding of the realities of pharmaceutical research. What he is describing here is removing many of the solid, evidentiary rules that are in place to make sure that the drugs reaching the market actually work and replacing them with proxy predictions. Look, my least favourite part of the job is the animal testing, and I would be hugely grateful if that could be eliminated from the drug discovery pipeline. People have been trying to do that for a long time. But it's extremely difficult. Biology is very, very complicated. Our understanding of how processes in organisms work are vague and approximative. This is not computer code. Even if Anthropic somehow got all of Big Pharma to hand them their proprietary data, it would only scratch the surface of the understanding that is needed to solve these kind of problems. Due to these realities, the program Amodei is describing here would, effectively, open a floodgate of drugs on the market that don't actually do what they are supposed to and are more likely to have unidentified toxicity.

It's not just due to energy, at least not directly. Core CPI (ex-food and energy) has been increasing monotonically since February:

https://fred.stlouisfed.org/series/CPILFENS#

There are tons of ways to generate "strong candidates for drug design." This is definitely not the bottleneck in drug discovery and development. The hard problem is vetting and developing these ideas to the point of having a commercially viable drug. That is still a very empirical process.