Screenshot of this question was making the rounds last week. But this article covers testing against all the well-known models out there.
Also includes outtakes on the ‘reasoning’ models.
Screenshot of this question was making the rounds last week. But this article covers testing against all the well-known models out there.
Also includes outtakes on the ‘reasoning’ models.
What worries me is the consistency test, where they ask the same thing ten times and get opposite answers.
One of the really important properties of computers is that they are massively repeatable, which makes debugging possible by re-running the code. But as soon as you include an AI API in the code, you cease being able to reason about the outcome. And there will be the temptation to say “must have been the AI” instead of doing the legwork to track down the actual bug.
I think we’re heading for a period of serious software instability.
It’s also the case that people are mostly consistent.
Take a question like “how long would it take to drive from here to [nearby city]”. You’d expect that someone’s answer to that question would be pretty consistent day-to-day. If you asked someone else, you might get a different answer, but you’d also expect that answer to be pretty consistent. If you asked someone that same question a week later and got a very different answer, you’d strongly suspect that they were making the answer up on the spot but pretending to know so they didn’t look stupid or something.
Part of what bothers me about LLMs is that they give that same sense of bullshitting answers while trying to cover that they don’t know. You know that if you ask the question again, or phrase it slightly differently, you might get a completely different answer.
This is necessary for sounding like reasonable language and an inherent reason for “hallucinations”. If it didn’t have variation it would inevitably output the same answer to any input.
AI chatbots come with randomization enabled by default. Even if you completely disable it (as another reply mentions, “temperature” can be controlled), you can change a single letter and get a totally different and wrong result too. It’s an unfixable “feature” of the chatbot system
This is adjustable via temperature. It is set low on chatbots, causing the answers to be more random. It’s set higher on code assistants to make things more deterministic.
Changing the amount of randomness still results in enough randomness to be random.
Yeah, software is already not as deterministic as I’d like. I’ve encountered several bugs in my career where erroneous behavior would only show up if uninitialized memory happened to have “the wrong” values – not zero values, and not the fences that the debugger might try to use. And, mocking or stubbing remote API calls is another way replicable behavior evades realization.
Having “AI” make a control flow decision is just insane. Especially even the most sophisticated LLMs are just not fit to task.
What we need is more proved-correct programs via some marriage of proof assistants and CompCert (or another verified compiler pipeline), not more vague specifications and ad-hoc implementations that happen to escape into production.
But, I’m very biased (I’m sure “AI” has “stolen” my IP, and “AI” is coming for my (programming) job(s).), and quite unimpressed with the “AI” models I’ve interacted with especially in areas I’m an expert in, but also in areas where I’m not an expert for am very interested and capable of doing any sort of critical verification.