I continue to think that unsupervised learning on car video data (next frame prediction) may be the best way to get true self-driving cars.
Pinned Tweet
Replying to
4
3
11
Follow
Rogs 
@ESRogs
Curious optimist.
Sincerity over sarcasm.
Believer in innovation (long $TSLA).
The only thing we have to fear is Moloch.
lesswrong.com/users/esrogs
Rogs 🔍’s Tweets
Bill Gates tries to install Movie Maker
January 15, 2003
257
3,386
24K
Show this thread
Wow, it looks like most of what I knew about Japan is outdated.
Japan builds a lot, has reasonably good fertility, and has increased immigration.
noahpinion.substack.com/p/actually-jap
9
23
145
“For 75 years, CDC and public health have been preparing for COVID-19, and in our big moment, our performance did not reliably meet expectations”
4
6
12
We're releasing our hackathon 1st place project - "GPT is all you need for backend" with
But let me first explain how it works:
73
409
1,659
Show this thread
hear me out: underwater high speed rail btw SFO & OAK
- 2 trains each way (BART to BART or after TSA checkin)
- 5 mins from SFO to OAK
- treat OAK/SFO as the same airport
- fund w/ $5 tkt surcharge on 18m flyers/year = $90m/y (1 mile of high speed rail = $200m, pay it off in 3y)
141
100
2,173
Very interesting summary of a couple of new papers from DeepMind:
1
Does anyone use "reinforcement learning from compiler feedback" to train LLMs for code generation? It seems like eg codex was just made by doing extra training on github code.
2
2
4
Show this thread
Mountain View is so back
93
350
3,342
WWI and WWII were very, very bad. This is Germany's population by age and gender in 1946
10
9
206
The future of email with ChatGPT.
163
3,764
20.3K
Show this thread
On February 22nd at 1.30PM, history will be made. For the first time ever, a robot will represent someone in a US courtroom.
DoNotPay A.I will whisper in someone's ear exactly what to say. We will release the results and share more after it happens. Wish us luck!
206
871
4,389
Show this thread
"If the AI gets an A on at least 5 of out 6 of those exams using same grading scale as his students, then Bryan owes Matthew $500. Otherwise, Matthew owes Bryan $500. For the purpose of this bet, an A- counts as an A."
15
12
69
No big deal; just a guy playing fetch with a beluga whale... 😮
235
5,263
28.6K
the first paragraph in Drexler's "Nanosystems" is by far the hardest-going introduction I've ever set my eyes on
25
90
771
Show this thread
Delighted to share with you our latest paper. Here, we test the hypothesis that aging may be driven by DNA damage-induced changes to the epigenome. Mightbe due to a glitch in the software of the body that causes it to malfunction, which can be fixed with a reboot?👇 1/
280
2,180
7,338
Show this thread
How to make ChatGPT 100x better at solving math, science, and engineering problems for real?
Teach it to use the Wolfram language.
ChatGPT: the best neural reasoning engine.
Mathematica: the best symbolic reasoning engine.
I can’t think of a happier marriage. 🧵 with example:
70
768
4,720
Show this thread
Just thinking about how the United States could have been the world leader in 5G technology instead of China if we had just given *one guy* a green card when he needed one
677
3,926
18.9K
Show this thread
My 22 year old cousin met his dream girl at a bar and it's going pretty well
5,776
74.2K
786.7K
Show this thread
Researchers trained a quadruped robot to learn how to walk directly in the physical world without simulators. Learning from scratch in only 1 hour was possible by using the Dreamer algorithm to continuously learn a world model and plan inside of it.
Link: danijar.com/project/daydre
60
429
1,753
There are cities that build lots of housing and where housing is cheap (green).
There are cities that build little housing and are expensive (purple).
There is no city in America that builds lots of housing and is also expensive. That giant red box is completely empty.
99
1,167
5,081
Show this thread
Talking to art director friend at a major studio. He reviews job applicants, mentions that his duties now entail identifying and eliminating AI art portfolios.
Vangelis plays quietly as I take a vape pull.
"Have you ever rejected a human artist by mistake?"
86
1,317
9,351
Show this thread
Introducing Cursor!! (cursor.so)
A brand new IDE build from the ground up with LLMs. Watch us use Cursor to ship new features blazingly fast.
87
381
2,648
Show this thread
When Murray Rothbard supported open borders:
11
18
121
Economics mini lesson:
Banning non-competes makes it harder for firms to justify investing in training
Who does this harm? Workers.
9
9
54
Show this thread
Financial market traders can choose to focus their info-collection efforts on “slow” info, that long stays relevant, or on “fast” info, that is quickly forgotten. My analysis suggests: such traders do seem to focus too much on fast info, relative to slow. overcomingbias.com/2023/01/fin-mk
2
6
21
$TSLA GROSS MARGIN IN Q1? did some hardcore math, backed into some great data, and what he found was quite shocking…
54
152
871
2022 was Numerai’s best year ever.
Numbers are now up on numerai.fund including new metrics like our Sortino Ratio.
20
23
170
We train Transformers to encode algorithms in their weights, such as sorting, counting, and balancing parentheses from lots of data.
I never thought we may also go in the *reverse* direction: *compile* Transformer weights directly from explicit code! Cool paper :
1/🧵
42
463
2,388
Show this thread
Twitter is the perfect environment to build an agentic language model: there’s a small action space, the main action (tweeting) is an NLG task, and there’s a readily available feedback signal to use for training (likes, retweets, etc)
7
5
84
Show this thread
on second thought, they may have Rube Goldberg’d this a little hard
confused on the difference between FCPs and Microsoft principal repayment here (are FCPs = pre-MSFT investors???) and also a few other things
Quote Tweet
1
1
6
Show this thread
this makes a lot more sense
Quote Tweet
Replying to @xlr8harder
looks like a much worse deal for OpenAI than the recent hype would suggest they could get. I'm not sure how the $29B valuation is calculated - $10B for 49% of a company would be $20B, and the profit deal would lower that significantly. I guess it's from the $1B they put in before
1
2
23
Real interest rates are interesting here because --
Real rates are high when
(1) growth is high, or
(2) probability of death is high
SO: real rates would be high if markets were expecting
(1) aligned AI (explosive growth), or
(2) unaligned AI (existential risk)
1
3
29
Show this thread
remember when we were all really impressed by those manually justified TXT files?
i just learned about 's justified TXT file that's also
AN EXECUTABLE x86 BINARY
not a shell script. exe binary. that happens to also be printable ascii.
3
2
63
Show this thread
Here’s the recipe to make Siri/Alexa 10x better:
1. Whisper to convert speech to text. Best open-source speech model out there.
2. ChatGPT to generate smart home API calls and/or text response.
3. VALL-E to synthesize speech. It can mimic anyone’s voice sample!
Quick figure 1/3
101
1,114
5,331
Show this thread
A thing that my views have evolved on is that I used to think that we'd see machine intelligence blow past human-level competence without slowing down, because human intelligence is not special. I still believe that human intelligence isn't special, but...
1
1
7
Show this thread
Anthropic's imminent release of a plausible GPT-3.5 competitor has huge implications for the space
Breaks OpenAI monopoly and demonstrates feasibility of technology for other industry entrants
10/10 news for startups and builders in the application layer
Quote Tweet
Compare to GPT-3, Claude (a new model from @AnthropicAI) has much more to say for itself.
Specifically, it's able to eloquently demonstrate awareness of what it is, who its creators are, and what principles informed its own design:
Show this thread
8
39
251
This is stated tendentiously but honestly I agree it’s weird that people can’t just admit they are OK with doing something morally suboptimal. You don’t gotta attack people for seeming to try harder than you in some area
EA hate has the same dynamic
(I’m not vegan or very EA)
Quote Tweet
Vegan hate is just cope from people who don't have the resolve to do what is obviously morally correct but also don't have the strength to admit that to themselves or others.
Show this thread
4
1
25
this is the single best interaction i've had with our assistant
23
136
631
Show this thread