Joel Grus 🤠 (@joelgrus) | teapawt

Pinned Tweet

I don't want to generate hallucinations from any model that uses water or copyrighted training data, or plagiarize with any model that hallucinates or uses water, or use water to run a model that does hallucinations or plagiarism. You know, as a career, I don't want to do that. – at Scenic Oaks, TX

2

15

14,802

"it says here you like to attend 'AI Safety' conferences?" "yes, but only for the sex cult stuff" "oh ok good"

5

390

my experience traveling in Europe is that when you tell people you're from the US they usually ask what part, and then they get excited when it's somewhere they've heard of like Texas or New York or California

On a group tour and the guide asks everyone where they’re from, absolutely everyone says the name of the country they’re from except the Americans who say their state ????????????

1

11

943

writing about the AI discourse all like

12

435

Joel Grus 🤠 retweeted

David Cournapeau @cournape

Replying to @joelgrus

It is surprising to me it surprises people working in data science. That has always been the number one issue of most DS projects I've been on. That blog post from a while ago explained it well: siliconcontinent.com/p/what-…

What do consultants get paid for?

The analysis is the easy part

siliconcontinent.com

1

6

420

Joel Grus 🤠 retweeted

The Blue Cross/Blue Shield Association says that AI is already driving *up* healthcare costs, as hospitals use the technology to find instances where they can bill more for the same level of care. bcbs.com/about-us/associatio…

166

4,403

877,981

you know who else had the ability to tell a moral and ethical story about how we live together

Former President Obama says Democrats made a mistake by making goals materialistic and instrumental: "What people are starved for at the moment is the ability to tell a moral and ethical story about how we live together in this moment."

4

642

which is why we need more administrators to oversee "faculty honorifics"

Max Kozlov @maxdkozlov

SCOOP: Top researchers at Columbia, Stanford, and others thought they were elected to elite scientific societies alongside Nobel laureates. The catch? The organizations are a sham. I spent the last month investigating a massive academic scam, for @Nature nature.com/articles/d41586-0…

Exclusive: Sham scientific societies are misleading star researchers

Nature - A Nature investigation reveals a network of organizations that have been building their ranks through deceptive practices.

9

1,244

one reason I am still an AI skeptic is *organizational illegibility* it doesn't matter how big your context window is if all the necessary context is tacit

7

49

2,564

"If You Fine Tune A Language Model Is It Still A Language Model" - the greatest thread in the history of forums, locked by a moderator after 12,239 pages of heated debate,

3

10

1,093

according to the obscure-word vocabulary test everyone is sharing, I'm sort of a midwit at least I'm still pretty good at math (until someone invents a test for that too)

1

8

685

tfw you don't know about map-reduce

Matthew Yglesias

How do you review petabytes of agent activity logs? That’s a lot of logs.

5

114

9,717

gpt-2: parrot gpt-3: also a parrot gpt-3.5-turbo: more of a macaw gpt-4o: parrot-esque gpt-5: um, can we put the whole parrot idea behind us already? gpt-6: who ever said that AI models are parrots? certainly not me!

7

53

2,111

"what do you call a parrot with a harness and tools?" "not a parrot"

1

9

621

Joel Grus 🤠 retweeted

Replying to @MelMitchell1 @BlancheMinerva

how does "RL post training" change anything, you still end up with a stochastic "predict the next word" model and the harness is basically a while loop and a bunch of tools, how does that eliminate "parrot-dom"

6

42

1,776

I was just talking about this yesterday For the average computer user how is any software (Office? Gmail? Amazon? Zoom? Slack? Spotify? etc) better than it was a couple of years ago ("Excel now has a copilot button!")

Lewis Campbell @LewisCTech

I still can't get over the complete lack of impact agentic coding has had on software from a user's perspective. Our industry looks faintly ridiculous for waxing lyrical about the "agentic era" when it's had zero observable impact to the end user.

68

1,368

42,538

relevant

Steve Jobs would fire the entire company and replace them with AI if he was alive to see this

1

93

4,328

today in my LinkedIn app the "more" link doesn't do anything

1

42

2,459

wait until they find out about "Battle of the Sexes"

Rachel Cohen Booth

this is weird and not funny, you should not have called it this

1

32

10,502

Joel Grus 🤠 retweeted

Kieran Klaassen

@kieranklaassen

I've been using Jev for all kinds of things. This morning I had a realization I kind of like: Use it to make non-black-box embeddings. Instead of an embedding model spitting out 1,536 numbers that mean nothing, you ask Jev questions about each document. The answers become the vector. An email in Cora: "I got charged twice this month, pls fix asap" [is_customer, urgent, about_billing, needs_reply] [1.0, 0.9, 1.0, 1.0] A newsletter: [0.0, 0.0, 0.0, 0.1] A friend asking about lunch: [0.0, 0.1, 0.0, 0.7] Then it's just old-school cosine similarity search. Search "billing issues from customers" as [1, 0.5, 1, 0.5] and the double charge comes out on top. Same idea for our articles at Every: [is_tutorial, about_ai, contrarian, beginner_friendly] Or support tickets: [is_bug, angry, churn_risk, enterprise] Every number has a name, so you can see why something matched. Need a new dimension? Add a question. Want urgent stuff first? Change the query vector. Trying this in @CoraComputer now to make search fast.

73

982

141,583

here's the slides for my talk "The Last Mile of AI Is Longer Than You Think" at the Optimized AI Virtual Conference a few weeks ago (talk track in speaker notes as always) docs.google.com/presentation…

16

947

guy who keeps tryinou tg different phone keyboards hoping that one of them will make hi a competent tyoid tsn (me) (I am very good / fast at typing on a computer keyboard!)

2

632