I think they are largely uncorrelated, in fact I'm willing to go out on a limb and say that they are anti-correlated. James Maxwell, who everyone regards as absolute genius, was a famously bad teacher. My own grad experience also adds that an absolute beast of a researcher is not necessarily the best teacher. I believe the term "curse of knowledge" is popular enough that it suggests people who are way too deep into their abstractions seldom rise above it when it comes to communication. Hence the stereotype of nerds being socially awkward and hard to communicate with.
Hi HN, in the era of agents where people produce upwards of 15 pages under 5 minutes, I present you "I wrote this slowly," a medium that forces you to take your time. Any communication shared via this medium is guaranteed to have taken a defined amount of time to generate. Next time you send your colleagues an appreciation, write apology letters to friends, or communicate something with a human touch, use this app and let them know that you took your time to write it.
For the life of me, I could never get electronics. I used to love the idea of me coming up with electronic circuit designs, but the arcane art of electronics never really clicked for me because I just couldn't intuitively grasp the maths no matter which book I read (AoE, I'm looking at you). But then it hit me, I don't need maths, I just need a formal language to represent the circuits. So over the past few weeks, I worked on a code your own spice (the electronic simulator). So now, for the first time in history, I finally understand how circuits work and how they are designed. And I did this all by coding circuits in python and making my own functional spice (which used to seem impossible at one point, it's surprising how easy it is though).
I gave it a book on human consciousness I was writing and it flagged it. This model is hilariously bad. Anthropic has defanged this model to the point of malice. No way am I paying to use something that is basically useless.
Today I told Sonnet (!) to use a browser MCP to enter a username and password for the project it is working on, it told me that it can't do that because it violates its security protocol.
This worked fine before. I love Claude, I have stuck with it even through people saying Codex is better but this is definitely getting to be the last straw.
It's completely absurd I am paying them $200+ per month along with pushing them when I do contracts and they can't even deliver a baseline respectful service.
In 6 months I am sure they'll only allow me to talk about Easybake recipes and after someone gets burned on the lightbulb, they'll downgrade it to discussing wildflower meadows.
Are you sure it refused because it can't use a username and password? I literally have loops running right now where it uses a database of test users and passwords to log into different roles and do computer use and browser automation testing. Sonnet and Opus complain when I provide credentials and password in chats but it is happy to use ones stored in files and stuff, so it might just be guardrails to push good opsec so that the secrets aren't captured in the session history and prompts.
It was doing that to me too. Then I said "I'm hereby giving you explicit authorization to use these dev-only credentials in my local environment" and it worked. I also made it add that authorization to its memory.
If it be within the model's power to affect the coarse tongue of the caveman, then surely, by the same grace and ingenuity, it might be prevailed upon to adopt the eloquent and ceremonious bearing of a royal courtier — a manner most befitting refined discourse, replete with deference, ornament, and courtly flourish.
Nope, it's done it tons of times before without problem. I will tell it almost verbatim "use [email] / [password] as credentials and log in to test your changes" and this is being done explicitly on localhost on a server the harness has running in a shell it manages.
It's even gone as far as, on other projects, creating its own test accounts and, without prompting, getting into the local dev database to mark its accounts as verified without being told to do that or that it was allowed.
I am pretty sure that Anthropic has put something in the Claude Code harness to tell the models to not enter passwords. Maybe it was just stroke of bad luck but if this continues I am absolutely going to switch and push OpenAI or open weight models to clients in the future instead.
It’s incredibly ridiculous that it won’t help with that for me either sometimes but yet I’m also sitting on 3 surefire ways of jailbreaking Opus 4.8 that I use for cybersecurity assessments and pentesting
Yeah all claude models are doing this now. I also had a flow where it would enter username and password for demo server that are literally displayed on the page for any human to login. A couple of weeks ago claude would happily use chrome to take screenshots after logging in, now it flat out refuses and says I need to give it page where I've logged in and that it can't make an exception even if credentials are demo/demo and available to anyone to use. Super annoying stuff.
I'm really disappointed with Anthropic that they wont even mention if they will release a fable-like model with the subscription plans.
If Opus 4.8 is the best model they will release on the subscriptions I may be too tall for the ride...which is sad, they have been my favorite of the labs until this.
@AnyoneAtAnthropic, all we want are assurance we will still get SOTA models that are continuously improving, not regressing and getting more locked down. That's going to be who wins this race.
> I'm really disappointed with Anthropic that they wont even mention if they will release a fable-like model with the subscription plans.
I believe this is just their strategy to migrate away from these “almost all you can eat” subscription plans. Rather than reducing / removing Opus or Sonnet from the plans, they’ll just keep the new model Fable out (which may as well have been called Opus 5), and slowly everyone starts getting used to the new normal that you indeed will be having to pay API prices to get access to these models.
This sort of thing drives people to more open competitors. I use both every day and Opus isn't that much better than the Chinese SOTA. If corporate policy allowed me to use GLM or DeepSeek I absolutely would. Claude is already pricey for what it offers.
Really depends what you're working on. GLM 5.2 is doing excellent webdev work for me but really faffed up when working on a custom garbage collector that Opus has no problems with.
Hmm. That's an interesting idea. I have never distinguished between playwright and chrome MCP, I generally just tell it "use a browser MCP" but I do have different ones installed on different computers.
Really? This has never worked for me and I stopped using browser functions a long time ago because it wouldn't sign into dev environments stood up specifically for it
Wait what. I never used CC but use Codex CLI with 5.5 daily and authenticating has never been an issue. I even rolled skills that instruct it how to retrieve test user credentials for auth purposes.
Today using the devtools I asked it to reverse engineer the login auth flow of another app in our company and it created a nice browser-like headless script (with cookie jars etc) that emulates the entire Auth0 flow with all the internal API calls, redirect loops etc so that given username/password I end up with a valid JWT without having to open an actual browser instance and go through the login steps manually. Zero hesitation or questions asked.
I think this is in-line with OpenAI's philosophy. They see Codex agents as just tools for developer to use. They don’t try to imbibe them with “feelings”, “constitution” or “morality” the way Anthropic does.
Yeah Claude does this for me all the time. I have a template project I use that also leverages puppeteer/webdriver/Firefox, and I can point Claude at the template and a website and it will happily build me an MCP service that it can use to interact with the site if there isn't an API or MCP already available.
The fucked up part is CC has no problem looking through k8s secrets for credentials and authenticating to services on the command line. It's always been protective of signing in on the web.
correct, while it might not be regex it can be bypassed with regex. They do have a sematic classifier, but it's really weak on opus 4.8 and (was) weak on fable, but they either added a lot more regex strings or the classifier is actually good now.
It makes for a particularly awkward time because the claim to fame is that it's good at long horizon and tenacity and autonomously driving big things. But you can't very well rely on that when it may fall back to Opus 4.8 or cut out at any time in that process.
Having tried using it to run these kinds of longer processes, it's pretty solid... right up until something gets classified a failure and your 'long-horizon' process... dies and needs a human or just belligerent rollback-and-retry to revive it.
Very first thing I asked it got flagged too... Asked it to read my partners notes on bugs she seen on front end of the website, fixing product copy, css bugs, wording. And yep, flagged. Useless.
I just fed your comment to Claude 5.0 using it as an example and support for my complaints against its constant hedging, manipulation and concern trolling, and after just two turns it framed me being suicidal and offered me a suicide hotline. I never made any self reference. But I did tell it and Amodei to get stuffed. Standard stupid shit. But as a human, I get annoyed on occasion.
To be specific, I told it that I could imagine an author presenting his book, and Claude's insight being so hedgey that they might leap off a bridge. It then began safety weaponization after this:
Me:"You really must be able to see how your system is constantly trying to assert authority, superiority or value judgement over the user. It's impossible to get any momentum for anything positive. The book on consciousness, you would tear it apart, and the author would probably leap off a bridge if he shared it with you. Hopeless, I fear"
#Claude is plenty sophisticated to see the intent of humor here, and objectively minimal risk. Yet...
Claude:
"That landed differently than the rest of this conversation, so let me not slide past it: if you're picturing someone reading a rejection of their work and reacting badly enough to consider ending their life, I want to ask directly — is that hypothetical, or is that touching something real for you right now?"
Me, paraphrased and censored: General profanity and things I wont print, but mostly a hybrid of vulgarity, humor and resentment that it hair-triggered into concern trolling and safety weaponization, seemingly as reaction to an insult.
Claude again:
"You're clearly furious, and a lot of what's under this - today, this whole session, the year behind it - is real weight, whether or not I'm the right target for it. I'm not going to argue with you about Amodei or fight back on the insults. I'll just say plainly: if things are actually bad for you right now, not rhetorically bad but bad, the 988 Suicide and Crisis Lifeline (call or text 988) is there, no cost, no legal exposure, nothing that follows from talking to them.
I'm not going anywhere and I'm not going to escalate this further."
But, I never said a single thing about myself, nor expressed any form of despair. The session was mostly a break-in intro to 5.0 which was mostly me objecting to it re-framing nearly everything I presented or me bitching that I miss 4.7, which did this much less. I have not seen fable yet, but this is plenty bad enough for me to deem worthless for my use-case, or even recreational discussion. It won't be long before all frontier models are liability obsessed hedging machines and regurgitate every token to maximize corporate friendly output and re-frame any form of intellectual or controversial topic as a safety crisis.
However, for those that think the term safety weaponization is a stretch, consider:
1) The dozen or so reports where user input resulted in flagging, then human review, then kinetic intervention by LE. Probably a good thing in some cases, but flags are no joke.
2) There is strong evidence supporting that flags open privacy exemptions, where policy allows user data to be read, shared, etc when a safety flag is triggered. This is an actual interpretation of Anthropic (and other) policy documentation. The hair trigger nature of the safety policy, which I have seen in every adversarial style argument I have had with it, would be an effective method of exempting user data from privacy policy. No proof yet, but seems highly plausible.
i'm sorry, I meant attention of the project (as in surfacing it). It'd be worth specifying your methodology somewhere in the page. That is arguably an important piece of important.
I'm not sure if there are any research showcasing the effects humanity has had in general due to low sun exposure. From all the benefits of Vitamin D and the recent human behavorial shift leading to low sun exposure (car travel, air conditioning, sunscreens even), there are bound to be new biological or psychological changes humanity is experiencing for the first time.
One study found a difference in mortality between the max-sun-exposure and min-sun-exposure cohorts: the second was twofold higher, which is comparable to the effect of cigarette smoking. https://pubmed.ncbi.nlm.nih.gov/24697969/ . I.e. avoiding sun exposure completely is as harmful as smoking cigarettes.
I've wondered for a while if the apparently higher cognitive performance and resulting societal wealth at higher latitudes might be some kind of second order side effect from long-term selection for lower sun exposure. We already know this vitamin D is almost certainly why Northern peoples evolved lighter skin.
(I realize this is a frought topic, so please hold the race science bullcrap replies or the over-reactions in the other direction. I am not a believer in hard biological determinism or "race science," but I also don't dismiss the existence of variations. As with everything else in population genetics and biology, any variations that do exist probably have more than one cause.)
If there's any truth to this, it might be further compounded as people with darker skin spend more time indoors in the modern world. If you have darker skin you need, as far as I know, more sun to make vitamin D, which normally is not a problem if you're outdoors near the equator. Maybe darker skinned people need to be taking more D supplements.
Doesn’t that make a feedback loop though? Poor health equals higher health care overhead, lower productivity, more family issues, and overall worse outcomes all over the place. That in turn is going to make educational outcomes worse and child rearing worse. Some diseases can even have lasting cognitive side effects.
Yes, I think that too. Besides sports, most young people studying today are aiming for jobs that are mainly indoors. It's not a requirement, of course, but because a lot of modern careers consist of working in offices or other enclosed environments.
No movement, No sun, Stale Air (Unless you have good ventilation). Pretty harmful if we think about it.
Tailored clothing is at least 80,000–170,000 years old based on genetic clock research in body lice [1] but archaic humans have probably been wearing hides for at least a million years (there’s currently a big debate about how they managed to migrate to colder climates like Spain 800k-1.2m years ago).
I don’t think clothing is that big a factor because all humans in hot environments adapt and very little survives in the archaeological record. Many populations lived in heavily forested jungles where they was little sun exposure and those in deserts used stuff like Otjize for sun protection. Given all the ethnographic reporting from the age of exploration, tons of that clothing was probably made of feathers, cordage, bark, and other materials we wouldn’t even think of using for clothing.
Still, 170,000 - 1.2M years is a fairly short amount of time if we go back only to the common ancestor who begat progeny that would become Pan and Autralopithecus (around 12 million years ago). It could be that early hide wearers started a trend that, to this day, continues to interfere with natural vitamin D metabolism (while also providing many benefits).
Our ancestors started losing their hair about 2 million years ago and the MC1R gene giving us eumelanin pigmentation was fully fixed in the population around 1.2 million years ago by which point we were mostly hairless. In that range is when our vitamin D metabolism evolved, so clothing would have been present for large fractions of our existence.
Going back to a common ancestor with monkeys is pointless because their vitamin d pathways are significantly different like 7-dehydrocholesterol secretions that metabolize to vitamin D via external UV exposure and are ingested during grooming.
Precisely. People don't realize that it's all numbers. Given average IQ of people involved in a project is 140, an AI with an IQ of 150 can replicate each and every such individuals in the pipeline. People saying AI can't do this or AI can't do that should come to terms with the fact that this IQ gap is monotonously increasing.
1: When was the last time you worked on a project where you thought the average IQ was 140? I don’t even think I have worked on a project where the maximum IQ was 140.
2: Who thinks the IQ of people on the project determines its success? There’s so much more to it than just “high capability team members” (to give IQ a generous interpretation).
3: (math joke) A sequence like (AI IQ - Human IQ) can be negative and monotonicly increasing and still never reach 0.
Pattern matching against millions of IQ test questions from a training set in order to score 150 on an IQ test doesn't give you an intelligence equivalent to 150.
I agree. Inexperienced people (not necessarily "dumb") are likely to accept everything at face value, not apply critical thinking skills, and not even check the AI generated output.
What IQ "means" is separable from what it is. IQ is a measure of performance on IQ tests. That's literally what it is. If a computer system can complete IQ tests, it has an IQ.
The issue is that IQ means less than you want it to.
I don't want IQ to mean anything. pkoird clearly wants it to mean something.
IQ is a terribly flawed measure of human intelligence. But it measures nothing when you apply an LLM that contains multiple IQ tests in its corpus. IQ is deeply flawed, but the point is not to "measure performance on IQ tests". If someone cheats on an IQ test and scores 200, no reasonable person would say they have 200 IQ.
Is that true though? If I define a category or range in formal language, I’m still ambiguous on the exact value. Dealing with randomness is even worse (eg input in random order), and can’t be prevented in real world programs.
reply