John Gordon · @jgordon
324 followers · 5731 posts · Server appdot.net

I recall when some were reassured by the math limitations of LLMs.

“It allows the AI to do math (very complex math) and do more accurate work with words (like actually counting words in a paragraph), since it can write Python code to address the natural weaknesses of Large Language Models in math and language.“

oneusefulthing.org/p/what-ai-c

#ai #llm #beafraid

Last updated 1 year ago

Real_JamesCain · @real_jamescain
1048 followers · 5542 posts · Server techhub.social

Anyone getting serious / vibes from this rotisserie?

#Cronenberg #goldblum #beafraid

Last updated 1 year ago

Jody Lemoine 🇨🇦 · @ghostinthenet
345 followers · 710 posts · Server hachyderm.io

Healthcare and financial services may make more use of AI.

#beafraid #beveryafraid #tfdx

Last updated 1 year ago

JJ Celery · @jjcelery
299 followers · 2125 posts · Server mastodon.ie

Despite my tiredness I decided to finish slip-stitching the right cuff after all.

Can confirm: yes, I do have the focus, commitment, and sheer will to stab something a thousand times.

#sewing #beafraid #beveryafraid

Last updated 1 year ago

Happy Donkey · @t3rrybowman
191 followers · 573 posts · Server mindly.social

It's almost like democracy is a block of ice slowly melting in the late winter sun, with a squadron of fascist tinkers chipping away at the edges with hammer, sickel, chisel and awl.

#1984 #BraveNewWorld

#lifeimitatesart #wakeup #beafraid #handmaidstale #fahrenheit451 #animalfarm

Last updated 2 years ago

John Gordon · @jgordon
268 followers · 3826 posts · Server appdot.net

GPT-4 gets a B on final exam for Aaronson's Introduction to Quantum Information Science. The wolf is here.

scottaaronson.blog/?p=7209

Without doing the course. About average for the brilliant undergrads who dare this class. Likely would do better with Mathematica plug-in.

Also got an A in Bryan Caplan's Labor Econ midterm. After losing his bet Caplan said: "AI enthusiasts have cried wolf for decades. GPT-4 is the wolf. I’ve seen it with my own eyes."

#ai #chatgpt4 #beafraid #jgshare

Last updated 2 years ago

John Gordon · @jgordon
261 followers · 3705 posts · Server appdot.net

I see threads about how our testing measures the wrong things because C4* does better on the tests than the vast majority of humans.

This is known as begging the question. (en.wikipedia.org/wiki/Begging_). Also, motivated reasoning.

*ChatGPT4

#chatgpt4 #ai #beafraid

Last updated 2 years ago

lady troodle · @TriciaScott
62 followers · 545 posts · Server toad.social
John Gordon · @jgordon
251 followers · 3398 posts · Server appdot.net

ChatGPT's 155 IQ and passing the Sebastian's children puzzle

notes.kateva.org/2023/03/chatg

I laughed and laughed but it wasn't a happy laugh.

#beafraid #chatgpt #jgshare

Last updated 2 years ago

John Gordon · @jgordon
249 followers · 3343 posts · Server appdot.net

This is for all the people who are not afraid of ChatGPT 4.

A few weeks ago I wrote a blog post on tips for geriatric

Today I asked GPT to write a similar post. It's kb is from fall 22 so it doesn't have my post.

You can see both mine and ChatGPT here:

notes.kateva.org/2023/02/tips-

Fuck but it's good.

#crossfit #ai #chatgpt4 #beafraid #whitewater

Last updated 2 years ago

John Gordon · @jgordon
247 followers · 3252 posts · Server appdot.net

I figured that would be another “warning” that was about something manageable (bad search), but he’s got my fears from a slightly different angle.

I think we are . HOWEVER, in 1945 I would have bet on nuclear catastrophe by 1960. My prediction accuracy is not great.

social.paulkedrosky.com/@paul/

#fucked #beafraid #ai

Last updated 2 years ago

John Gordon · @jgordon
248 followers · 3211 posts · Server appdot.net

GitHub Copilot X: AI assisted coding.

mjtsai.com/blog/2023/03/23/git

CGPT 4 helps build the code to build CGPT 5. Don’t worry, it’s not a positive feedback loop.

#beafraid #jgshare

Last updated 2 years ago

John Gordon · @jgordon
245 followers · 3018 posts · Server appdot.net

I ask ChatGPT about the next step: NLP and domain-specific knowledge modules.

notes.kateva.org/2023/03/i-ask

A better answer than 99% of the online comments on ChatGPT.

But don’t worry. It will be fine.

#ai #beafraid #buckleupbuttercup #jgshare

Last updated 2 years ago

John Gordon · @jgordon
234 followers · 2818 posts · Server appdot.net

Gassée has the SV investor take on GenAI.

mondaynote.com/genai-more-nets

It’s very bullish. His list of things to fear is much too short.

#ai #beafraid

Last updated 2 years ago

John Gordon · @jgordon
231 followers · 2782 posts · Server appdot.net

Noah Simpson on Lower Haight’s AI Tech Culture and it’s antecedents.

noahpinion.substack.com/p/ai-t

Some of the best parts of his essay are descriptions of older tech cultures. I was there a few months ago and this explains some puzzles. The social scene reminds me of Caltech student houses in 70s.

Also, Noah is ancient for this crowd.

#ai #beafraid #jgshare

Last updated 2 years ago

John Gordon · @jgordon
231 followers · 2720 posts · Server appdot.net

In 2008 Aaronson said the was reassuringly far away.
scottaaronson.blog/?p=346

Today: “I apologize for anything I said before the deep learning revolution that was, on balance, overly dismissive of the scenario…” scottaaronson.blog/?p=7042

If you know who Aaronson is then you know Team got a big boost today. (Note - Aaronson is keen to talk with the God Emperor to come.)

PS. I like to think I had a bit part in his responding to his 2008 post.

@cstross

#ai #singularity #beafraid

Last updated 2 years ago

John Gordon · @jgordon
228 followers · 2686 posts · Server appdot.net

Google engineer Blake Lemoine - LaMDA AI “sentient”. June 2002.

washingtonpost.com/technology/

He was fired. His cognitive science background likely informed his definition of sentient. Many laughed at him. I did not laugh then, I do not laugh now.

#beafraid #ai #chatgpt #jgshare

Last updated 2 years ago

John Gordon · @jgordon
225 followers · 2593 posts · Server appdot.net

Note how this works. The language model “speaks” *then* the integrator “understands” and then the math module is invoked.

This may also be how some of us think. (Studies of mind suggest multiple flavors of consciousness. Famously some people have an internal narrator, others do not.)

appdot.net/@jgordon/1098987450

#ai #cognitivescience #chatgpt #beafraid

Last updated 2 years ago

John Gordon · @jgordon
214 followers · 2548 posts · Server appdot.net

@paul Welcome to Team . At least we won’t need to worry about climate change.

Also, what does the XP tag mean?

#beafraid

Last updated 2 years ago

JJ Celery · @jjcelery
223 followers · 1001 posts · Server mastodon.ie

On a further note, dying your own hair is as much self care as it is a crime scene cleanup job training.

Can't leave the bathtub looking like I've murdered someone in it, now can I.

#hairdye #beafraid #beveryafraid

Last updated 2 years ago