miraculixx

I like to think of LLMs as informed dice throwing

Close the PRs as unacceptable.

Who'd want to work in an environment like this? It's not sustainable

Stop accepting these PRs. Ask for a personal 1:1 explanation. Make it the submitter's problem, not yours.

So then they should get 10-100x more revenue, now that AI does all the marketing and selling.

Interesting! However there is a somewhat misleading statement in the readme where it says "no code required". Well the workflow specs are in fact code. Other than that I applaud the approach.

Did they compare it to other models? A lot of this sounds like this is the first time they have applied AI to security, and they are just amazed at the unreasonable performance of a pattern matching machine. Well, it matches patterns. duh

welcome to the club :) I came to the same conclusion a year ago and uninstalled all the AI assistants that my IDE tried to force on me. Back to good old auto complete and it works great. Feeling productive and on top of things.

So you continue to show ads to Copilt, just not to the user? If so, not a fix.

This is interesting. How do you keep this up to date so quickly?

I agree in general, but how are you ever upgrading any of that? Could be a "sleeper compromise" that only activates sometime in the future. Open problem.

Sure because it worked great when we tried last time, right? Just spec it out first, and AI will churn out the perfect app. Not.

For anyone who hasn't worked in a waterfall project and would like to try: You are kidding yourself.

There is no such thing as a perfect spec. Read that again and say it outloud.

It took humanity 50 years to figure out that perfect specs are impossible, unless of course you know exactly(!) what you need. And even then, the specs are never complete.

The reality is that we often don't know, and can't know, what we want, exactly, until we actually see and experience what we said we wanted. Then we adjust. Try again.

That's the reality for individuals already. By simply using logic we can deduct that entities made of more than one individual, aka companies, will not behave better. They just make it look better by giving you a nice document that says "we want this!", only to then come around when they see what they got, and to claim "wait, we didn't mean it like so!".

That's just human nature. Not much we can do. AI will not change that.

So when some people think AI will deliver perfect software given perfect specs, hence we have to write the specs first! That is just missing the boat my a mile.

Agile is not a process but a human-friendly way of doing things. It simply says hey you want this? Let me build that and show you. Then let's adjust or move to the next thing. Rinse and repeat. Agile works because it matches how humans think and act. Step by step, day by day.

can someone verify that?

Simon just explores stuff and writes about them. Doesn't mean he uses LLMs for everyting.Antirez likes to question stuff and make them better. Doesn't mean he uses LLMs for everything.

Also their experience is not my experience. I will make my own choices.

Just stick with pip and venv.

Windows is the root cause here, not pip

+1

+1

Anaconda solved the same problem ~10+ years ago already.

That is not correct. The EU AI Act has no such provision, ans the data mining excemption does not apply as the EU has made clear. As for Switzerland copyrighted material cannot be used unless licensed.

Are these 56 cases distinct from all other cases in the data?

Exactly. The study has been set up to produce this exact result. They essentially limited the human doctors to bare essentials, on specialist cases(!), while providing the LLMs with all sorts of help, including discussion among several AIs.

That's like letting one group of students have a strict closed-book exam, while another group can take the test as a group exercise and accessing any material they like, then claiming that closed-book exams lead to worse outcomes.

In a nutshell the study is just slop designed to get attention. The headline result is what they really want people to hear, and that's all the media will be repeating.

As any AI researcher knows, if you have a model that does 4x better than the naive baseline (the humans, in this case), you are likely looking at overfit, not real-life performance. This study is just slop, and you can tell so by the mere fact that they did not submit a paper, but just published a PR article.

To fly means "to soar through air; move through the air with wings" (etymonline)

That is pretty much an accurate discription of what planes and birds do.

To plan means "to reason with intent".

That is very much not what LLMs do, and the paper does not provide evidence to the contrary. Yet it uses the term to give credence to it's rather speculative interpretation of observed correlation as causation.

Interestingly enough there is no definition of the term, which at least would help to understand what the authors actually mean.

I would be more inclined to take a more positive stance to the paper if it used more appropriate terms, such as call observed correlations just that. Granted that would possibly make for much less of a fancy title.

Very much in support of this. The use of anthropmorphic or even biological terms are entirely misguided. All they do is drive a narrative that is very much belitting natural intelligence.

Unfortunately that's not a review but a hype-driving oversimplification of what the paper says.

"Hallucination" is just to term we use to say "this result is not what it should be". The model always uses the very same process, it does not do one thing for "hallucinations" and something else for "correct" results.

In a nutshell it is always predicting the next token from a joint probability distribution. That's it.

All other interpretations are speculative.

> This doesn't seem to happen very often in classical programming, does it?

Try concurrent programming. It happens all the time.

We know how they work. It's just that it works better than expected. Which of course doesn't mean we don't know, it just means there are second-order effects that are non-obvious. Intelligence is not implied.

There is no evidence to this end. There is lots of claims, but claims are not evidence.