I tried to create an "intuitive" explanation of LLMs without going into the math but still removing a little of the "black box" sense people get around them...
I think there's something about the way this is written that's endemic to our time -- it basically feels right because it gives a possible explanation of current events that has some internal consistency. It doesn't mean it's right, but neither the author nor (most of) the audience care about that part.
I feel like politics will be automated eventually. We have voters who express intents, then we have politicians who are (in theory) hired to best represent and interpret those intents.
But there's only 1 signal (election/not-election) and it's only delivered every few years. Yes, there's some voter feedback in the interim sometimes, but unless there's also a lot of press and noise around the voter feedback, politicians are unlikely to do anything with it because it probably won't impact re-election.
Ideally, there would be much more interactive feedback between voter and political agent (human or otherwise) carrying out voter intents, especially around informing of unintended repercussions of policy decisions and ultimately forcing voters to have more nuanced policy views.
At least that's what I hope for. What reality says is usually quite different.
"In the first year" is a good and well-needed qualifier for such analysis.
Oftentimes we fall into the trap of defining a business success as "forever", anything short of that is deemed a failure. I think such views make the industry look far bleaker than it actually is.
Even by their standards, I'm not sure that 'running a business that did well for 10 years' should count as a fail in the same way that 'couldn't even figure out a business plan' should.
I don't know if this is a myth that ought to be perpetuated.
The skilled CEOs I know are all exceptionally well-attuned to the problem space that their product solves, and their own company's place in that space (usually including internal company dynamics). Yes, from an employee's perspective, the CEO is pretty magical -- they usually have a better 'big-picture' view of the task you are trying to accomplish than you do, and they can give you the right context, validation, resources etc. to totally turn your work life around.
However, this is usually limited to the problem-space and the company itself. If you're in the same problem-space or dealing with similar company problems then a conversation can be quite valuable. But it doesn't sound like that's what you're looking for so much as inspiration.
You might get that from talking to a CEO. They can tell you the situation and circumstances that set them on their particular path, and maybe you can extrapolate nuggets of wisdom out of that. But they won't be able to figure out the path that's right for you. They're not oracles or mystics, just people with the right intersection of skills deeply specialized in the right cross-section of the market.
To me, it feels like it's started giving superficial responses and encouraging follow-up elsewhere -- I wouldn't be surprized if its prompt has changed to something to that effect.
Before, if I had an issue with a library or debugging issue, it would try to be helpful and walk me through potential issues, and ask me to 'let it know' if it worked or not. Now it will try to superficially diagnose the problem and then ask me to check the online community for help or continuously refer me to the maintainers rather than trying to figure it out.
Similarly, I had been using it to help me think through problems and issues from different perspectives (both business and personal) and it would take me in-depth through these. Now, again, it gives superficial answers and encourages going to external sources.
I think if you keep pressing in the right ways it'll eventually give in and help you as it did before, but I guess this will take quite a bit of prompting.
>To me, it feels like it's started giving superficial responses and encouraging follow-up elsewhere -- I wouldn't be surprized if its prompt has changed to something to that effect.
That's the vibe I've been getting. The responses feel a little cagier at times than they used to. I assume it's trying to limit hallucinations in order to increase public trust in the technology, and as a consequence it has been nerfed a little, but has changed along other dimensions that certain stakeholders likely care about.
Seems like the metric they're optimising for is reducing the number of bad answers, not the proportion of bad answers, and giving non-answers to a larger fraction of questions will achieve that.
I haven't noticed ChatGPT-4 to give worse answers overall recently, but I have noticed it refusing to answer more queries. I couldn't get it to cite case law, for example (inspired by that fool of a lawyer who couldn't be bothered to check citations).
> I think if you keep pressing in the right ways it'll eventually give in and help you as it did before, but I guess this will take quite a bit of prompting.
The rush to adopt LLMs for every kind of content production deserves scrutiny. Maybe for you it isn't "avoiding work" but there's countless anecdotes of it being used for that already.
Worse IMO is the potential increase in verbiage to wade theough. Whereas before somebody might have summarized a meeting with bullet points, now they can gild it with florid language that can hide errors, etc
I assume you're talking about ChatGPT and not GPT-4? You can craft your own prompt when calling GPT4 over API. Don't blame you though, the OP is also not clear if they are comparing Chat GPT powered by GPT3.5 or 4, or the models themselves.
I can understand why they might make changes to ChatGPT, but it seems weird they would "nerf" the API. What would be the incentive for OpenAI to do that?
The API though? That's mostly used by technical people and has the capability (supposedly) of querying different model versions, including the original GPT4 public release.
A huge difference between ChatGPT and crypto, though is that the latter literally offers to print money.
The former, what, lets you make low-quality internet content a little more easily? It's harder to share the long-term 'get-rich-quick' benefits (though, of course many are trying right now). There's an initial bump right now as various content mediums try to cope with ChatGPT content, but once that stabilizes, I don't see the long-con grift that the author is mentioning being possible in the same way.
K, I'll play. Let's say that Reinforcement Learners (algorithms/strategies/agents in Reinforcement Learning), let's say that they have some property of 'consciousness' that's similar to humans.
A 'reinforcement learner' gets positive or negative feedback and adjusts its strategy away from negative feedback and towards positive feedback. As humans, we have several analogs to this process.
One could be physical pain... if you put your hand on a stove, a whole slew of neural circuitry comes up to try and pull your hand away. Another could be physical pleasure, you get a massage and lean in to the pressure undoing the knots because it's pleasurable.
If we look at it from this angle, then if we're metaphorically taking the learner's hand and putting it continuously on the stove, this would be problematic. If we're giving it progressively less enjoyable massages, this would be a bit different.
Even more different still is the pain you feel from, say, setting up an experiment and finding your hypothesis is wrong. It 'hurts' in some ways (citation needed, but I think I've seen studies that show at least some of the same receptors fire during emotional pain as physical pain), but putting a human in a situation where they're continuously testing hypothesis is different from a situation where their hands are being continuously burned on a hot stove.
I think, then, that the problems (like they alluded to here) are:
- how can we confirm or deny there is some kind of subjective experience that the system 'feels'?
- if we can confirm it, how can we measure it against the 'stove' scenario or any other human analogue?
- if the above can be measure and it turns out to be a negative human scenario, can we move it to one of the other scenarios?
- even if it's a 'pleasurable' or arguably 'less painful' scenario, do we have any ethical right to create such scenarios and sentiences who experience them in the first place?
I think this argument would have to conclude that training any RL agent at all is unethical, since updating its weights 'away' from some stimulus could be considered pain.
I think what I'm getting at is that there are different types of 'pain', some that we'd consider ethical and some that we wouldn't.
Constantly subjecting a person (or some abstract simulation that responds like a person) to the equivalent of continuous bodily pain would be deeply unethical, but, say, giving a person clues towards solving a puzzle would be considered less so.
Overall, I think you're right though, if we somehow discovered we've created simulated people (or sentient beings) then we probably shouldn't use them to solve arbitrary problems.
RL agents are exposed to millions upon millions of stimuli, almost all of which will be 'painful' at least initially (according to the websites definition of pain). I think for children negative stimuli are not all painful, pain is a certain small subset of negative stimuli that damage is being done.
I think children are constantly experiencing rewards and punishments of a sort analagous to reinforcement learning. Until a certain age their brain is hardwired to want to please their parents, so they pay a lot of attention to the parents' faces, and find reward in smiles and find punishment in frowns. After that, they are driven by a certain amount of ego, and find reward in self-accomplishment and find punishment in being told what to do.
I've seen consultants hired for projects where everyone on the team (including leadership) basically knew what to do or what needed to be done, but they were only, say 75% certain about it. In a normal startup, they would've just done it, failed fast, moved on. But in a bigger corporate (non-startup) environment where they didn't know if their jobs would be on the line for taking risks, they could bring in consultants to verify but mostly echo back what they already wanted to do. Plus they come with certificates and stuff so everyone in the chain assumes that's worth something.
