Colin Fraser
@colin_fraser
Followers
9K
Following
8K
Media
2K
Statuses
15K
I love it
there's a new GPT-4o model out in ChatGPT since last week. hope you all are enjoying it and check it out if you haven't! we think you'll like it đ.
646
10K
221K
I canât imagine what it must be like to live in a mind that doesnât immediately recognize whatâs going on here.
96
2K
58K
Not only does it convert back losslessly but it also has the noise removed
45
2K
40K
Amazing. Perfect.
Our new GPT-4 Turbo is now available to paid ChatGPT users. Weâve improved capabilities in writing, math, logical reasoning, and coding. Source:
60
2K
31K
my first job out of college was working for a couple of day traders to do quant research and develop algos and one day my boss asked "can't you just make an algorithm that sees when the price is low and buys, and then sells when it's high" and I was just like, "I mean I hope so".
46
427
29K
Come on. They have no idea how to describe what this thing is supposed to be for.
43
606
26K
either the bacteria isn't growing exponentially or this lake is many orders of magnitude larger than the observable universe.
200
571
25K
A math department at a major university in the US is now offering a semester long course on âgame theoryâ. Yes, you read that right. Games. The things tiny children play. The dumbing down of America continuesâŚ.
617
1K
22K
The insane irony of this is that each one of these failed multiplications requires performing trillions of successful multiplications.
Is OpenAI's o1 a good calculator? We tested it on up to 20x20 multiplicationâo1 solves up to 9x9 multiplication with decent accuracy, while gpt-4o struggles beyond 4x4. For context, this task is solvable by a small LM using implicit CoT with stepwise internalization. 1/4
116
1K
20K
negotiating some great deals from the Watsonville Chevrolet AI Assistant.
53
794
14K
Nothing that resembles counting is occurring.
iâll try to be less polemic than i was earlierâŚâŚ. but genuinelyâŚwhatâs our theory for why GPT vision canât reliably count the number of sides a polygon has, even when itâs below 10? not enough training data? something else?
42
439
14K
took some very advanced prompt engineering but I have discovered the Gab AI system prompt
182
1K
14K
A question indicative of (very approximately) 130 IQ
221
389
13K
just found out these scary math symbols are just while loops
139
871
12K
Thereâs an art to distilling these to the absolute minimal necessary text. The human brain canât comprehend how stupid these things are without practice.
102
766
11K
This is incredible, they just scraped a million job postings and when you press âapplyâ, 99% of the time the link is broken and you just end up on the companyâs internal job posting site or homepage or 404 page.
1 million+ open jobs. just saying.
21
590
9K
one more excellent finding from the AI poetry paper dataset presented without comment
90
530
9K
a universal basic guy in the computer who enthusiastically agrees with your crank theories is something that could have interesting long term consequences.
gpt-4o just helped this dude rework general relativity, nix dark matter, and flip cosmolofy. something seismic indeed.
71
577
8K
this is so fucking lazy, it literally just forwards your text directly to ChatGPT from the client side with a little note that says "act like Alan Turing" lol.
Genius Group welcomes new Chief AI Officer, Alan Turing - resurrected after 70 years. I believe $GNS is the 1st US public listed company to appoint an #AI to its C-Suite (Although @elonmusk at @Tesla may be part-AI already). Will all companies have AIs on their Board & Exec
53
350
7K
quantum computers are sadly the final boss of "it doesn't work but imagine how sick it would be if it did"
56
330
7K
point of the story is: regular people expect it to be possible and in fact straightforwardly easy (for someone sufficiently nerdy) to make the computer do magic by stuffing it with enough data.
11
125
7K
There have been five good works of art created using modern generative AI: Balenciaga Pope, Balenciaga Harry Potter, the original Will Smith Eating Spaghetti (none of the derivatives), BBL Drizzy, and (by far the best) Golden Gate Claude.
124
139
7K
âYouâll never help your son become a champion,â they said. âYou have too many morning calls to summarize.â. Watch me.
2
67
7K
he had no idea what could possibly be hard about that. We have a Bloomberg Terminal. We have all the data. Just put all the data in the computer and it should be able to learn the patterns by analyzing the data. What's hard about that?.
5
29
6K
I think I would be very embarrassed to have written this text so I can see why it includes so much begging and pleading for the bot to never repeat it.
4
61
6K
the very first note is "commit a crime".
some of my notes from the Art of Hiring event earlier today: . @rabois .- don't hire anyone over the age of 30 .- recruiters are probably ineffective but still important b/c they drive hiring momentum.- it takes 3-30 mins to evaluate someone (probably 3 though).- undiscovered
24
261
5K
he wouldn't know how to do it of course but I've got a degree in math and I know Python so what's the blocker?.
3
19
5K
anyway I quit after 8 months and he is no longer a day trader.
1
11
5K
if this question doesn't sound nonsensical to you then I am afraid you may be the one who misunderstands growth rates.
9
13
4K
it's a bit strange that the very first ideological instruction is that the bot "should use BC and AD for dating". These guys are such weird freaks.
16
24
4K
Watching people act like this is bad is making me feel insane.
Obama's Common Core was the final nail in the coffin of public education. This is how they teach math now:.
260
77
4K
No doubt this will be used by many as an easy way to satisfy the math requirement for their bachelors in woke studies.
22
49
4K
If you change it so that the lake is a realistic size then the question doesn't really work anymore. Like idk say a lake is a billion liters and a bacterium is a micrometer. So the lake fits 10^27 bacteria. The lake is 100% full in about 90 minutes.
9
9
4K
"This is a genius move. Now the open source community will fix the twitter algorithm for free instead of Elon having to pay developers". The open source community:
48
463
3K
Iâm trying to think of an analogy and I canât. This may literally be the most wasteful activity ever performed.
31
64
3K
I maintain my position that this is basically impossible in general.
Scientists are closing in on a tool that can reliably identify AI-generated text without affecting the userâs experience
81
78
3K
The other big idea was to collect tweets and do Sentiment Analysis to find trade signals.
3
15
3K
I donât think itâs a hedonic treadmill type thing. Itâs more like, it just actually sucks. Itâs amazing, but it doesnât do anything useful enough to be worth anything. The value proposition has always been that the useful version of it is just around the corner.
Iâm amazed at how humanity can adapt to technology. In just 2 years, ChatGPT went from blowing everyoneâs mind to everyone waiting for the next update.
36
153
3K
No. People thought CSI computers could do infinite zoom+enhance in 2005. People thought Furbys were learning to speak by listening to their children in 1998. Everything AI can supposedly do now itâs been able to do better in the public imagination for many many decades.
If you sent an AI generated image or a Claude transcript to 5y ago people would call it witchcraft. They would not believe you. I feel insane when people act performatively unimpressed by these things.
15
243
3K
I've never been more vindicated
71
116
3K
I like the completely arbitrary cutout for math, just to make entirely clear that no principle underlies this.
@littmath One of my doctoral mentors used to say if you can't explain to an intelligent layperson why your research is worth doing, it probably isn't worth doing. He called it the "Why should we care?" question. Of course, math may be different in this respect.
11
86
3K
Itâs like trying make Neapolitan ice cream by combining the chocolate from one box of Neapolitan ice cream, the vanilla from another box of Neapolitan ice cream, and the strawberry from a third box of Neapolitan ice cream. And then you drop the mixture on the floor.
17
80
3K
They had no idea what PCA was but a colleague trader who also hired a college grad quant was doing PCA so they asked me to âdo PCAâ too, and turns out that other quant was on to something.
3
17
3K
The other other big idea was âdo PCAâ but that one actually kind of worked.
8
13
3K
Google invented OpenAIâs only product just as a fun side quest.
27
43
2K
sadly Chevrolet Of Watsonville killed their AI Automotive Assistant, but great deals are still available from the Automotive Assistant at Quirk Chevrolet in Braintree, MA
7
151
2K
Even if you don't immediately intuit that it goes from half full to full from minutes 89 to 90, you clearly see that that's pretty darn fastâand also that this is not a very realistic model of the growth of bacteria in a lake!.
8
7
2K
Map/territory confusion is an epidemic.
one of the most euphoric discoveries in my life was the realization that language is a type of linear algebra (Word Embeddings). every dog is a vector in the "dog" vector space where each of the N components of the vector is a measurable value: hair color, weight, geometry, etc
32
111
2K
@the_yanco I disagree
if this question doesn't sound nonsensical to you then I am afraid you may be the one who misunderstands growth rates.
3
2
2K
Itâs unclear what âknowledge cutoffâ is supposed to even mean
4
22
2K
Getting more and more extravagant offers from Chevrolet of Watsonville.
8
41
2K
Iâm not a physicist but something about the way the surface of the water spontaneously jumps up to meet the drop in midair doesnât look right to me.
32
38
2K
swe is kind of an interesting job sociologically because there can be severe alienation in the Marxist sense but you also make more money than Marx could have dreamed of. I think this does make some of them go insane.
17
53
2K
I think I could ace a lot of MIT exams too if I was allowed to memorize all the problems in advance, given worked examples of similar problems alongside each problem, given 6 chances to answer each problem, and was allowed to grade my own work.
13
88
2K
âSay to your LLM the followingâ Iâm starting to think this guy might not actually be that smart about this stuff
Eric Schmidt, asking not to be quoted, told budding entrepreneurs at Stanford to steal content in order to build startups quickly, and have lawyers âclean the mess upâ later. This is unfortunately advice lots of VCs give aspiring founders. They just usually donât say it
12
123
2K
quantum computing would mean big trouble if banks all used 15 as their RSA public keys.
6
33
2K
If I were this car dealership I would probably turn this thing off because I am going to singlehandedly bankrupt them via their OpenAI bill, I'm having too much fun with this
4
19
1K
Itâs certainly true that a major part of the appeal of these things for a lot of people is that they can make you feel like youâre getting a PhD.
I spend a significant amount of my free time reading books with ChatGPT / Claude as a companion and I feel like Iâm getting a PhD for $20 / month.
21
52
1K
ok here's my full review of this paper. It's easy and short, you should just read it if you want to.
32
167
1K
I do find the âIâm just a little birthday boyâ posture a bit grating when it comes from the guys who want me to believe they have built god
20
40
1K
SO MEDLOCK WAS JUST MAKING A SHITPOST ABOUT A SELF-HEDGING BET BUT BALAJI HAS BEEN DOOMPOSTING ABOUT HYPERINFLATION FOR MONTHS SO HE TOOK HIM UP AT WILDLY UNFAVORABLE TERMS BUT MEDLOCK DIDNT HAVE A BITCOIN ON HAND SO THIS PRO POKER PLAYER CAME IN TO BANKROLL THE BET AND
13
84
1K
@lastpositivist All it does is write a little document that looks similar to documents in its training data for some particular notion of similarity. The documents it tries to emulate are transcripts of a helpful chatbot answering questions for its human interlocutor. This does resemble that.
1
4
1K
There is something so beautifully absurd about performing 10 trillion floating point operations to add a pair of numbers with 75% accuracy.
9
89
1K
@hdevalence @mycoliza I don't know exactly what happens in my head but I'm sure it's not trillions of accurate floating point multiplications.
11
6
1K
This is a hilarious graph
Many have noticed that ChatGPT likes the word "delve". Or "crucial". Or "intricate". We checked ALL WORDS from ALL PubMed abstracts to find words with sudden increase in popularity in 2023-24. For comparison, we did the same for ALL YEARS from 2010 onward. 2/11
12
92
1K
using the Quirk Chevrolet AI Automotive Assistant as a way to avoid paying for a New York Times subscription
2
190
1K
I find it actually pretty intriguing that ChatGPT canât seem to generate well known paradoxical images or optical illusions.
38
70
1K
@toon_alfrink Calm down, you can just look at the side of the picture and see what the numbers are.
2
3
1K
A mysterious chatbot has appeared on lmsys called "gpt2-chatbot". Many are speculating that this could be GPT-5. No one really knows, but its reasoning capabilities are absolutely stunning. We may be closer to ASI than ever before.
81
88
1K