"LLM paragraph will register to much of your audience not as writing but as output"
The best advice for writing for other humans is: don't use LLMs.
If you're writing for machines, liking coding, then fine go for it. If you're writing for processes with formal highly structured content like manuals, specifications, form content, procedures, information, that sort of thing, then also ok to use LLMS. But if you're writing for a human mind to ingest and extract meaning from, then LLMs are poison.
Hallucinations don't really work that way. They happen when you ask an LLM to generate new material. Not so much reviewing existing material.
But there's always the "ask a second LLM to get two opinions" technique which gets you pretty close to error free.
And... now that I think of it... even if it was terrible and missed 50% of the errors... wouldn't that still be better than using no LLM which catches 0% of the errors?
LLMs exclusively hallucinate. That is the sole mechanism their reasoning is based on. The fact that they are ever correct is a happy coincidence.
You can reduce the likelihood of incorrectness by adding the real facts to the context, which is what LLMs do. But that in and of itself is not a guarantee that the LLM will be correct. Because its fundamental process of reasoning is probabilistic word-by-word hallucination.
And it would be an odd way to describe all the extremely useful things they discover. Are you suggesting they're taking "magic" mushrooms of some kind?
On the balance of probabilities, the CRISPR thing is a hack job. AI did not do it. It's likely the AI-assisted work of researchers white labeled as AI.
Do you just believe every bit of press release that you read? These companies have an enormous amount of pressure on them to make profit. They can and will lie to try and convince you that AI is better than it is.
Is there any evidence outside of these companies that AI is breaking new ground on literally anything? No. It can help a little with a lot of different types of work– none of it novel.
Personally I think every single medium to long form piece of writing should go through an LLM grammar check at the minimum (they are great at catching stuff like correcting “reign in” to “rein in”, not to mention their/they’re and so on) although I don’t think they generally have the self control to make the corrections themselves without screwing stuff up.
I can see being lazy and skipping it, but refusing to use them even for more minimal advice purposes seems crazy to me.
Eddy_Viscosity2 · · focus · HN ↗
The best advice for writing for other humans is: don't use LLMs.
If you're writing for machines, liking coding, then fine go for it. If you're writing for processes with formal highly structured content like manuals, specifications, form content, procedures, information, that sort of thing, then also ok to use LLMS. But if you're writing for a human mind to ingest and extract meaning from, then LLMs are poison.
qarl · · focus · HN ↗
Don't you think they're useful as fact checkers?
tenuousemphasis · · focus · HN ↗
qarl · · focus · HN ↗
Hallucinations don't really work that way. They happen when you ask an LLM to generate new material. Not so much reviewing existing material.
But there's always the "ask a second LLM to get two opinions" technique which gets you pretty close to error free.
And... now that I think of it... even if it was terrible and missed 50% of the errors... wouldn't that still be better than using no LLM which catches 0% of the errors?
devilsdata · · focus · HN ↗
You can reduce the likelihood of incorrectness by adding the real facts to the context, which is what LLMs do. But that in and of itself is not a guarantee that the LLM will be correct. Because its fundamental process of reasoning is probabilistic word-by-word hallucination.
qarl · · focus · HN ↗
Well - that's not how the term is used.
And it would be an odd way to describe all the extremely useful things they discover. Are you suggesting they're taking "magic" mushrooms of some kind?
devilsdata · · focus · HN ↗
qarl · · focus · HN ↗
Are you not aware?
devilsdata · · focus · HN ↗
Do you just believe every bit of press release that you read? These companies have an enormous amount of pressure on them to make profit. They can and will lie to try and convince you that AI is better than it is.
Is there any evidence outside of these companies that AI is breaking new ground on literally anything? No. It can help a little with a lot of different types of work– none of it novel.
qarl · · focus · HN ↗
My friend, I use AI every day. I am fully aware of how fucking amazing it is.
When you can no longer clamp your ears and eyes shut, come find me. I'll love the apology.
cheesecakegood · · focus · HN ↗
I can see being lazy and skipping it, but refusing to use them even for more minimal advice purposes seems crazy to me.