ChatGPT lies then gaslights reporter with fake transcript

[Jfdlsjfd]

I've heard this before and it always strikes me as odd. I'm genuinely curious: if asking, "Are you sure?" identifies most hallucinations, why isn't "Are you sure?" functionality being used under the hood to proactively stop hallucinations from happening?

Maybe that's just the difference between commercial AI application sold 300 €/user/month and the 20 $ solution for the general public

Also, it might be that answering the second question consumes as much resources as answering the first, behind the hood. That would push the entry price to a level many casual users wouldn't be ready to just try. I feel there are more people that can spend 20 USD a month on that work 95 times out of 100 for their use case, fails 4 times and fumbles critically 1-in-100, rather than pay 100 USD a month for a tool that would work 99 times, fail critically 0.2 times and fail regularly 0.8 times.

Honestly, while it can work, it's not totally enough: I have seen ChatGPT doubling down on a wrong answer about the ir'Wynarn family line in the Eberron setting. There was no official answer, fan sites gives conflicting information, and he hallucinated and kept doing it until asked, "OK, show your sources."

 

[The Firebird]

These practical uses don't require the massive money burning that more "commercial" models do.

It's somewhat inseparable though, right? It's hard to take an area of technological progress and say "we're going to see how it affects this useful thing but ignore everything else". NVIDIA spent decades optimizing GPUs for gamers, and one could have argued it was a waste of energy and brainpower to invest that much into a frivolous pursuit.

Does AlphaFold start using transformers if Google doesn't have the capital to fund research into transformers? Does Google care if there isn't a monetization opportunity for whoever solves translation?

 

[Umbran]

I would agree. As it stands I think that power production would be in danger of going in exactly the opposite direction, as that tech is already available to serve the "immediate (created) need."

Fair point.

I was more considering technology development - NASA uses public funds to develop a ton of tech, and in making the patents for those available returns more value to the economy than the agency costs. That's before we get into the value of support of satellite operations t our communications infrastructure, and so on.

AI Tech companies return value to their investors, not the economy as a whole, and they aren't even profitable at that, yet.

 

[Whizbang Dustyboots]

Yes. Now move on to the next step.

If large swatches of people literally cannot eat or live because of the situation, do you think the status quo stands?

Seriously, you've correctly predicted the a likely case, now predict what comes next.

A sea change is coming.

I've been to actual warzones and slums the size of whole cities in the west.

Even if you hate the current (very hateable) status quo, I would caution against getting too excited about the sea change.

 

[Jfdlsjfd]

Fair point.

I was more considering technology development - NASA uses public funds to develop a ton of tech, and in making the patents for those available returns more value to the economy than the agency costs. That's before we get into the value of support of satellite operations t our communications infrastructure, and so on.

AI Tech companies return value to their investors, not the economy as a whole, and they aren't even profitable at that, yet.

Honestly, some of the leading research is done by state-owned universities, EU and China are providing computing grants on public funds, and tax returns on research make France (the 35 hour workweek, 2 hours lunch break when you're not on strike "investor-hell") the first destination of FDI in Europe. This is quite a similar strategy of public support.

In some sectors like image/video generation, Tencent and Alibaba (or Mistral for LLM) are open sourcing their models. They don't seem to be looking for returns at this point. It is very possible that private funding isn't a sustainable model for this technology in the long run.

 

[kermit4karate]

I've heard this before and it always strikes me as odd. I'm genuinely curious: if asking, "Are you sure?" identifies most hallucinations, why isn't "Are you sure?" functionality being used under the hood to proactively stop hallucinations from happening?

Good question. My guess? Cost. The basic models are good enough for most queries and uses. When I ask ChatGPT to be careful before responding and to doublecheck its work, it even tells me it's switching over into "Thinking longer" mode.

Thinking longer mode requires more cycles to compute, costing OpenAI more.

 

[Scribe]

Any human who screws up will get it brought up again and again just like we are doing with AI, but I'm talking in the limited case of a human artist vs generating images.

I mean sure, people still dunk on Liefeld so yes, I guess people will continue to comment about individual arts vs the monolith that is AI generated images.

I'm not sure how its relevant to you, the AI doesnt have feelings, but sure.

 

[Jfdlsjfd]

Good question. My guess? Cost. The basic models are good enough for most queries and uses. When I ask ChatGPT to be careful before responding and to doublecheck its work, it even tells me it's switching over into "Thinking longer" mode.

Thinking longer mode requires more cycles to compute, costing OpenAI more.

And if you use it too much, it is (or was at a time) hard-limited. I had a game-book like experience that consumed my monthly allotment of "Thinking mode" but it didn't suffer a lot of the common problem often linked with AI.

 

[kermit4karate]

Directly below the text box for ChatGPT where it says "Ask anything," it states, "ChatGPT can make mistakes. Check important info."

That's there for every query.

They might not be marketing it to you in specific. Given that you seem to be an early adopter. Most generally, there's a heavy push by companies that sell these services - and internally by the companies to justify spending so much on them- so that the average office drone uses theses tools and learns to love and rely on them because the average office drone gets easily frustrated by them and just doesn't see the point.

I see that. To those easily frustrated office drones (your phrasing), I'd offer that I'd love to help them craft better prompts! There are easy things you can do to enhance ChatGPT's truthfulness and veracity.

 

[kermit4karate]

And if you use it too much, it is (or was at a time) hard-limited. I had a game-book like experience that consumed my monthly allotment of "Thinking mode" but it didn't suffer a lot of the common problem often linked with AI.

Yup. But you can get more Thinking Mode for the low, low price of $200/month now!