ChatGPT at Age Two, an analysis of what has and hasn’t changed and where the field is still stuck. Short holiday read, w Ezra Klein cameo. open.substack.com/pub/garymarc... - ThreadSky

garymarcus.bsky.social • 87 days ago

ChatGPT at Age Two, an analysis of what has and hasn’t changed and where the field is still stuck.

Short holiday read, w Ezra Klein cameo.

https://open.substack.com/pub/garymarcus/p/chatgpt-at-age-two

Comments

hssndeatomized.bsky.social•86 days ago

3/final. This is based on speaking to people across a lot of fields, from architects, engineers, coders take your pick and also experience building things at work. You (the royal you) do yourself a disservice if tou ignore this reality.

hssndeatomized.bsky.social•86 days ago

2. For these GenAI tools most users have figured out (as you and others have been pointing out) that these things are just autocomplete on (super) steroids. Most users now realize this and use it exactly in this way, to iterate much much faster with humans firmly in the loop.

valoisdubins.bsky.social•86 days ago

Do most users know this?

I don’t think so. People repeatedly say it’s unreliable - but only in their own specialism.

hssndeatomized.bsky.social•86 days ago

Hmmm and maybe instead of "most' it should say "a lot", and I instead of "users", "users using these tools in anger to get stuff done"

hssndeatomized.bsky.social•86 days ago

The fact that they are autocomplete etc.? Likely they don't have the statistical picture, but they treat it as such in the sense that they know it will give you an answer but you can't trust it and the answer needs checking and fixing

valoisdubins.bsky.social•86 days ago

The problem is when the answer sounds convincing - which ChatGPT is very good at.

You and I may keep fixed in mind how it works, but a lot of people who should know better think LLMs have actually almost achieved true intelligence - “just wait till next year” etc.

hssndeatomized.bsky.social•86 days ago

Oh yeah that's extremely frustrating - I do not know what those folks are thinking. I don't have a full sense of how much of that has sunk into the consciousness of the general public. I think/hope not too much... but more worrying is when it influences gov policy.

hssndeatomized.bsky.social•86 days ago

1. I think there are two separate issues here: is GenAI (and the like) the right way forward for improving AI in the long term and are tools like ChatGPT useful? IMO the answer to the first (in total agreement with you) is a resounding no. The answer to the second IMO is a resounding yes.

richargh.de•87 days ago

> And what unnerved me a bit about ChatGPT was the sense that we are going to drive the cost of bullshit to zero when we have not driven the cost of truthful or accurate or knowledge advancing information lower at all.

richargh.de•87 days ago

> These systems have no conception of truth. Sometimes they land on it and sometimes they don’t, but they’re all fundamentally bullshitting in the sense that they’re just saying stuff that other people have said and trying to maximize the probability of that.

Powerful words

lennonjohn.bsky.social•86 days ago

Where is the killer app?

For electricity we can show the light bulb. Or power an MRI

For nuclear tech we can power a city or submarine. Or destroy the world.

Automobiles can deliver patients rapidly to a hospital so they DONT DIE

What will be on AI’s resume?

lipstickscribbles.bsky.social•87 days ago

The high priests of AI will never listen because they are interested, at their core, in manifesting a god.

A bunch of 64 bit Hubbards without the creative fiction skills.

When their temple falls, that's when we will usage for the leftover blocks.

wordmonk.bsky.social•87 days ago

But it makes my bullshit look like filet mignon.

sanmei.bsky.social•86 days ago

As my dictionary and thesaurus sites of choice have begun using LLMs which perform the same work at a fraction of the efficiency or reliability, I've begun to have doubts that LLMs are even good at word associations - which should, I would think, be their actual best purpose!

jamesmarinero.bsky.social•86 days ago

As my Dad used to say 'Bullshit baffles brains'. I use Gemini for research and find it very helpful, but I crosscheck its output. I gave up on ChatGPT.

Comments

Posting Rules

Reply