The root cause of what happened in that story was ultimately uncontextualized question asking.
Basically this guy starts with this fringe conspiracy theory belief that chloride ions are bad for you and asks a question to Chatgpt about alternatives to chloride ions and gets bromide as the next halogen.
We don't know this for certain, but when that video came out I tried it in ChatGPT and it this is what I could replicate about chloride bromide recommendations. It doesn't suggest eating sodium bromide but it will tell you bromide can fit where chloride is. The paper that discusses the case also mentions this.
> However, when we asked ChatGPT 3.5 what chloride can be replaced with, we also produced a response that included bromide. Though the reply stated that context matters, it did not provide a specific health warning, nor did it inquire about why we wanted to know, as we presume a medical professional would do. [0]
Of course this kind of bad question asking makes you fall short of the no free lunch theorem / XY Problem. Like if I ask you: "what is the best metal? Name one only." and you suggest "steel" then I reveal that actually I needed to conduct electricity so that is a terrible option.
Yes I understand that the context matters and to be honest, the person's context wasn't really given publically but still, i think that they trusted AI sources itself and that confusion almost costed their life.
> We don't know this for certain, but when that video came out I tried it in ChatGPT and it this is what I could replicate about chloride bromide recommendations. It doesn't suggest eating sodium bromide but it will tell you bromide can fit where chloride is. The paper that discusses the case also mentions this.
From the video I watched, what I can gather is the fact that somehow the chatbot confused chloride and bromide for washing machine related tasks but that being said, AI's are still sycophantic and we all probably know this.
> Basically this guy starts with this fringe conspiracy theory belief that chloride ions are bad for you
I still feel like AI/LLM's definitely tried to give into that conspiracy rheotoric and the Guy got even more convinced as proof
Of course he had a disillusioned theory in the first place but I still believe that a partial blame can still be placed and this is the crux of the argument actually, dont read just AI sources and treat them as gospel
They are based on scraping public resources which could be wrong (we all saw the countless screenshots floating on internet where google search engine's AI feature gave unhinged answers, I don't use google so I don't know if they still do but for a time they definitely did)
This is actually what I think the grandparent of the comment is talking about regarding poisoning of data I think in their own manner or atleast bring the nuance of that discussion.
It might not be a conspiracy theory. Europeans have serious media skeletons in there closet.
Consider La Liga in Spain. When football matches are on they have a blank check to block whatever they want wherever they want. Genuinely they take down all of cloudflare and all kinds of shit. I think they were even DNS banning everyone on .tv TLD. Its wild how much legal power they have.
This was brought up on hacker news often.
They also have their apps spy on users microphones and gps to detect where someone might be watching their streams to make sure you aren't doing it in bars. [1]
Italian media is trying to do similar stuff with their piracy shield stuff. [2]
AtomicDig219303 on Reddit when Italy blocked all of google drive.
> Wait, I don't think that your post describes how fucking idiotic this whole thing is. Piracy shield is a system implemented by AGCOM (which as OP said is a governing agency) and basically "gifted" to the fucking mafia that is Serie A (yes, the football/soccer league) to block access to pirated streams of football matches.
NVIDIAs 8xB200 gets you 30ktps on Deepseek 671B at maximum utilization thats 1 trillion tokens per year. At a dollar per million tokens that's $1 million.
The hardware costs around $500k.
Now ideal throughput is unlikely, so let's say your get half that. It's still 500B tokens per year.
Gemini 3 Flash is like $3/million tokens and I assume it's a fair bit bigger, maybe 1 to 2T parameters. I can sort of see how you can get this to work with margins as the AI companies repeated assert.
Cool, that potential 5x cost improvement just got delivered this year. A company can continue running the previous generation until EOL, or take a hit by writing off the residual value - either way they’ll have a mixed cost model that puts their token cost somewhere in the middle between previous and current gens.
Also, you’re missing material capex and opex costs from a DC perspective. Certain inputs exhibit diseconomies of scale when your demand outstrips market capacity. You do notice electricity cost is rising and companies are chomping at the bit to build out more power plants, right?
Again, I ran the numbers for simplicity’s sake to show it’s not clear cut that these models are profitable. “I can sort of see how you can get this to work” agrees with exactly what I said: it’s unclear, certainly not a slam dunk.
Especially when you factor in all the other real-world costs.
Google runs everything on their tpus which are substantially less costly than to make and use less energy to run. While I'm sure openai and others are bleeding money by subsidizing things, I'm not entirely sure that's true for Google (despite it actually being easier if they wanted to).
I suppose this is a more forward looking post though. It's about whoever gets to whatever is awaiting us in the future 7 months before the other.
And I guess the idea is is that there is this extreme inflection point in utility somewhere that makes it so getting there first gives you some incredible economic edge.
It might not exist though. Like either utility plateaus and its bubble crash of the century time or it just keeps going up but without any specific point where you can differentiate something.
The GP is talking about recursive self-improvement.
What yes, it's clear by now it's way beyond the capacity of those AIs, and the odds are pretty good it's impossible to a large extent (but some limited version of it may be possible).
> We're just choosing between fraudster shop owners getting a cut, or the addict being able to buy twice as much malt liquor.
I don't agree with these zero friction in a vacuum takes. Difficulty in access does shape choices, a lot in fact.
If you make it easier for people to use handouts to gamble or do drugs or whatever then more people will do it and ones doing it will do more of it. This isn't even a take its the null hypothesis.
The null hypothesis could just as easily be if they get a 1:1 dollar exchange rate versus a 1:2 rate on their food stamps, they can afford to buy drugs AND food instead of just drugs. Guess which one they buy if they can only buy one? Guess what they are incentivized to do if they have less cash than they need on hand to do both? I'll give you a hint, it rhymes with teal.
I think that it should be a ramping rate, the idea being that a 1x landlord should be able to outbid a 2x landlord and so on.
In theory this encourages a sort of spreading effect where at some point the Nth property is too expensive to buy to rent or speculate on, which naturally stops the exponential effect of making land lording your full time gig by continuously expanding the portfolio.
Individual owners drive far more NIMBY policy than corporations. The rent of land exclusivity should accrue to the public rather than to private individuals or corporations.
This is not surprising at all. I'm not sure why people keep making this mistake thinking that good 'ol Republicans are flipping their script all of a sudden. MAGA is a populist movement and has a lot in common with the far left types.
NPUs are just kind of weird and difficult to develop for and integration is usually done poorly.
Some useful applications do exist. Particularly grammar checkers and I think windows recall could be useful. But we don't currently have these designed well such that it makes sense.
A while ago I tried to figure out which APIs use the NPU and it was confusing to say the least.
They have something called the Windows Copilot Runtime but that seems to be a blanket label and from their announcement I couldn't really figure out how the NPU ties into it. It seems like the NPU is used if it's there but isn't necessary for most things.
Basically this guy starts with this fringe conspiracy theory belief that chloride ions are bad for you and asks a question to Chatgpt about alternatives to chloride ions and gets bromide as the next halogen.
We don't know this for certain, but when that video came out I tried it in ChatGPT and it this is what I could replicate about chloride bromide recommendations. It doesn't suggest eating sodium bromide but it will tell you bromide can fit where chloride is. The paper that discusses the case also mentions this.
> However, when we asked ChatGPT 3.5 what chloride can be replaced with, we also produced a response that included bromide. Though the reply stated that context matters, it did not provide a specific health warning, nor did it inquire about why we wanted to know, as we presume a medical professional would do. [0]
Of course this kind of bad question asking makes you fall short of the no free lunch theorem / XY Problem. Like if I ask you: "what is the best metal? Name one only." and you suggest "steel" then I reveal that actually I needed to conduct electricity so that is a terrible option.
[0] https://www.acpjournals.org/doi/10.7326/aimcc.2024.1260
reply