I'm getting a lot of refusals these days from multiple LLMs on multiple fronts for silly stuff, a lot more than I had for a while. If this is where things are really going, I think open weight models have a big future.
That I don't know if they're a stranger or not. I introduce myself to acquaintances fairly regularly (sometimes annoying them that I apparently think they're so unmemorable) because I'm the opposite of 'I never forget a face'.
I think normal people are more likely to have the experience where they can't remember a name and why/how/where they know someone from. I of course forget things like anyone, but that's unrelated.
Like the poster, I'm faceblind. It isn't the worst thing: I'm not voice blind, height blind, age blind, hairstyle blind, gender blind, features associated with race and ethnicity blind, attractiveness blind, affect blind, context blind, etc., so I'm mostly good at figuring out who someone is. Within one encounter with a bunch of people, I try to note what someone is wearing.
Every once in a while I don't recognize someone and I go through this whole thing of bringing up every biographical detail about them I remember and all the things we've talked about to show that I'm not an asshole who wasn't paying attention in the past. Fortunately, I have a decent memory for such things.
LLMs are great with minority languages compared to almost anything else. Including better than the by the natural language generation employed to use Abstract Wikipedia, which whiffs at relatively large languages like Zulu and Xhosa, let alone many of the rarer languages that popular LLMs speak fluently.
Other correctly point out it does matter what language the code is in since the human does sometimes need to read and understand it.
But also, I suspect the article is just wrong. "The hard languages got easy first" isn't true in practice and the impressive examples given are not representative or as magical as the poster makes them out to be.
The takeaway might be right in the end, but the post isn't right in the beginning.
> Physicians use all their senses. They poke, they prod, they manipulate, they look, listen, and smell.
Sometimes. Sometimes they practice by text or phone.
> They’re also good at extracting information in a way that (at least currently) sycophantic LLMs don’t replicate.
If I had to guess, I think I'd guess that mainstream LLM chatbots are better at getting honest and applicable medical histories than most doctors. People are less likely to lie/hide/prevaricate and get more time with the person.
Would be interested to hear a legal expert weigh in on what 'advice' is. I'm not clear that discussing medical and legal issues with you is necessarily providing advice.
One of the things I respected OpenAI for at the release of ChatGPT was not trying to prevent these topics. My employer at the time had a cutting-edge internal LLM chatbot for a which was post-trained to avoid them, something I think they were forced to be braver about in their public release because of the competitive landscape.
I'm struggling to understand what the result really is: it seems that some dogs at some point would rather play with a toy than eat or come play with their owner. That seems pretty normal. Is this really "addictive-like"? Why isn't it "really enjoy"?
Whenever I try to read up on it, it seems like glaciers are receding at ~2x their without-climate-change rate. That's a huge increase, but it doesn't seem like there's something that a person can experience at a visceral level here that is based on fact and not just preconception.
It's definitely striking, I can't deny that. I crossed the last remnants of an almost-extinct glacier last year that my guide guessed would be gone in 1-3 years: at the beginning of his career it was a real glacer with non-trivial extents, crevasses, etc.
I make my x's with a backwards c and a c, like Computer Mondern and lots of fonts https://i.ibb.co/8LPsJKsj/image.png - doesn't look much like a chi or a times sign
No one in the target audience is using × for scalar multiplication.
The new management found many billions of assets of various asset classes https://www.reuters.com/technology/bankrupt-crypto-exchange-... When FTX thought it was broke, there were really billions of dollars of stuff they were too much of a mess to come up with.
That isn't to say that 'all the money was there' or that the conduct wasn't deeply criminal and unethical, but a huge part of the shortfall was really just terrible books.
The famous John Ray (emergency CEO) quote from a court filing was
> Never in my career have I seen such a complete failure of corporate controls and such a complete absence of trustworthy financial information as occurred here.
It seems like for your theory to be better, it should fit the facts better (not saying it doesn't). Whether it seems agentic or not doesn't seem very relevant to whether it's what happened.
You could also think of it as 11 minutes each for defrauding each of 1.2MM people. I don't know whether either of those sentences are good ones, but there does come a point at which a narrow violent crime should be punished less than a huge non-violent one. FEMA uses about 7.5MM for a human life for certain cost-benefit analyses (much lower values are used in many of the countries where most FTX customers lived): is there a sense in which doing a billion dollars of damage is in certain ways comparable to 130 killings? Surely some people who lost money would have used that money in part to save someone's life.
