It's not only you, there's an explosion of ChatGPT on HN

Can you provide some examples of the so many tasks that these LLM’s are performing for you that a human was needed before?

Regardless, crypto bros are the worst and I hate how every bull run we see an avalanche of new coins that all claim to be the solution to all our problems when they are really just a Ponzi scheme. Bitcoin and Ethereum are the only two that I feel are worth investing in long-term, but that’s just me.

Did they? It's the best performing asset in the last decade. I think they got it right.

Careful not to overfit ;)

Neat tool, thanks for sharing.

To me, there is no greater toy than a programming language. ChatGPT is neat, but it isn't a programming language. Same for blockchain or agile or whatever other trend is happening in the industry. Some of the trends literally are new programming languages! Golang is really fun! So is TypeScript!

Some trends or technologies make programming even more fun (in my opinion anyway) and I embrace those feverishly: distributed version control, CI/CD, pair programming (sometimes it's even more fun with a friend!), configurable linters like Perl::Critic, Intellisense, JUnit-style testing frameworks. All this stuff helps me feel more in control of the computer (or distributed cluster of computers), which I've discovered is the main thing that gets me off about programming.

I'm even still hopeful that LLMs will have some role to play in my having more fun with programming. I've tried CoPilot and so far it hasn't grabbed me, but maybe this will change. In any case, there are clearly other people having fun with it so I guess that's good. Maybe somebody can find joy by debugging GPT-4 prompts the same way I enjoy pouring over stack traces.

Hype cycle topics seem to attract opinionated folks who have strong feelings on topics and have the need to shout them from the rooftops, whether that's doomer, booster, or luddite. That's what I find exhausting.

I’ll also say, I think LLMs are different than bitcoin. It has its own killer app, and it has tremendous social impact, not necessarily positive. If anything, crypto doesn’t really make sense without AIs.

One thing though is that the foundational models are created and controlled by big tech. It’s been compared to silicon fabs and its scales of economy … However, I don’t see a future where foundational models can only be created by large orgs with a lot of resources is something beneficial for society.

The difference is I'm mostly tired of the social garbage that keeps piling up on top of things. The competitiveness, the pressure to "be productive", the grifters, the capitalists who put money before people, the bureaucrats. It's never enough more more more faster faster faster meanwhile the roadblocks that are put in place get bigger uglier and stickier than ever.

I just stopped participating in that stuff. Got off Facebook. Got off Twitter. Curated my Reddit feeds to be built around useful and helpful communities, not reactionary meme-ified BS. Started reading more. Started using RSS again (but again a reduced and focused subset of useful sites).

I'm feeling much better and I still find I enjoy technology. Microelectronics, 3d printing, functional programming, distributed systems.

I'm working on a cloud connected garage door opener. I don't care that it's never going to be productized. I don't care that I can go an Alibaba and order one for $15. I'm just doing it because I want to work on my microelectronics skills and I find it fulfilling. I'm doing it my way, at my own pace, without the BS.

And we thought there was an epidemic of bullshit jobs before! Not only will everyone be spending their days churning out this garbage, but everyone will be forced to read it as well!

This is not the dystopia I had in mind! It’s much more boring and missing the cool mirrored shades.

You're effectively saying that the "truth" problem with AI is solved by letting it guess. Which is probably a fine enough answer, but doesn't actually "solve" the problem at all.

To the points:

3) Training on data sets where the answer is never "I don't know" is effectively the same as raising children to believe they're always right and teaching them to be needlessly confident. No one should trust those folks.

4) "Lack of access to information" is not the issue... These models are trained on the entire corpus of Twitter, Wikipedia, and more information than I've ever seen in my lifetime, and some of them already do this (Bing) and produce little more than summaries of blog posts based on keywords. If anything, the issue is that an LLM lacks the real world knowledge to discern any nuance as to whether something is correct or grounded in reality.

I've now taken the time to watch John's talk and I have some thoughts. It's not only difficult to solve truthfulness due to disagreement (and subjectivity), but it's also very difficult because different contexts have different standards of evidence.

In some contexts, like programming, we'd rather have the model output its best guess for what the program should be, no matter how low the confidence, because we would like a starting point and we can debug the program from there. The answer "I don't know how to write that program" is not a useful starting point and it may even be an example of the model withholding information it does have due to low confidence.

In other contexts, such as scientific or historical questions, we want a high standard of evidence. Asking the question "what year did Neil Armstrong land on Mars?" should not produce a hallucinated response with fully unhedged language complete with fictitious date of landing. This problem may be solvable by training the model to hedge or even to question the premise when the confidence is low. Of course, this also suffers from the garbage-in-garbage-out problem of having falsehoods buried in the training set.

A more subtle and difficult problem with scientific/historical questions is with long-form answers. Currently, models tend to produce long-form answers that fairly consistently contain a mixture of true facts and falsehoods, and it can be quite difficult for even expert readers to spot all of the mistakes every time. Furthermore, the human labellers were given very sophisticated tools for highlighting sentences in long-form output but the information this produced had to be reduced down to a single bit per example since the detailed information did not improve training very much.

Personally, I think it's going to be very difficult to teach the model how to recognize the appropriate contexts and associated standards. This is a very subtle problem and one of the issues is that it relies on information the model does not have access to, for example: the identity of the question-asker. If a child asks an astrophysicist about black holes they're going to get a different answer than if an undergraduate student asks the same question in class. Yes, this additional context can be included in the prompt, but at some point it becomes a pain to have to copy-and-paste the context for every prompt.

Perhaps people will create a tool to save this additional context in the form of presets but this imposes additional effort on humans. At some point I think the amount of human curating and feedback that goes into these models will cause a collapse and backlash. We saw the same thing happen in the early days of search engines, when Google (fully automated) trounced Yahoo (human curated), leading to Yahoo's abandonment of human curation. We also see the same problem manifest itself at the Patent Office, where human review is policy. The entire patent system has become grossly dysfunctional at least partly due to the overwhelming complexity of this problem.

One thing I really liked was the "inner monologue" of the model performing a sequence of steps to answer a question by doing a search. If this could be generalized to other tasks it could be a home run for automated assistants (Google/Alex/Siri).