Zvi Mowshowitz
@TheZvi
Followers
28K
Following
3K
Media
133
Statuses
15K
Blogger world modeling, now mostly AI and AI x-risk, at Don't Worry About the Vase (https://t.co/tvn3lNcwc3 on SS/WP, LW), founding Balsa Research to fix policy.
New York City
Joined May 2009
Why do we think Claude is having such an epic fail at reaching consumers? Is it fixable?.
openai is winning consumer. how do i know?. my dad just asked me how to download chatgpt. most people haven't even hard of claude or perplexity
460
22
1K
The Jones Act protects us from this port strike, by ensuring that none of our domestic supply lines rely in any way on our ports. Our congress is wise indeed.
12
57
944
Today: .Wife: Have you heard about [person we know]?.Me: No. Wife: If I told you he had 50 million dollars, how would you think he got it?.Me (guessing): He hacked a crypto exchange?.Wife: You are correct.
4
37
863
@AmandaAskell Not my top priority or anything but I'm tired of being told I'm asking fascinating questions and raising interesting points and so on. I want to be able to trust such statements as meaning something.
24
10
773
You miss 100% of the shots you don't take, so yes I am interested in this OpenAI CEO role.
9
25
689
TIL that the Chinese call red teaming 'blue teaming' and I love it so much.
15
15
675
*Pulls up NVID premarket, sees it -11% on news its chips are highly useful.*. The Efficient Market Hypothesis is false.
41
42
1K
My lord, calm down, if this is true then you probably won't exist for much longer after that either way.
If you are a software engineer who’s three years into your career: quit now. there is not a single job in CS anymore. it's over. this field won't exist in 1.5 years.
13
22
633
It's. true? I've had 10+ quick and painless experiences in a row, at some point I might have to admit they're not the problem anymore.
62
3
551
Amazing news. So many of the most important consequences of this election, in all directions, are quickly revealing themselves to be things I did not hear mentioned even once the entire campaign.
BREAKING: Members of President-elect Donald Trump's transition team have told advisers they plan to make a federal framework for fully self-driving vehicles one of the Transportation Department's priorities, easing U.S. rules for self-driving cars.
24
39
517
Key question: How could the board be sufficiently motivated to move in this way with no notice or consultation with anyone, and also be open to discussions to resigning and bringing Altman back? Especially if this was motivated by safety concerns.
59
19
486
No way. You're kidding. He didn't say what I think he did, did he?.
13
14
494
OpenAI: The Battle of the Board. I highly recommend not reading it this way, go to my SS, but if you insist, this one plays less awfully than usual in pure text form with no links, so here you go:. On Friday afternoon, OpenAI’s board fired CEO Sam Altman. Overnight, an.
40
81
476
It seems this is real? In all seriousness, did we just. beat China?.
One of the most striking charts this year:. China’s startup ecosystem has almost completely collapsed in the last 5 years.
74
6
465
This week in Magic: The Gathering: The old card frame is awesome and we should go back to using it.
24
19
416
I outright predict that if an AI did escape onto the internet, get a server and a crypto income, no one would do much of anything about it.
@ESYudkowsky "It is only doing it in a few percent of cases."."Sure, but it is not agential enough to actually succeed."."Sure, it found the escape hatch but it was fake."."Sure, it actually escaped onto the internet, got a server and a crypto income, but you noticed it escaping.".
31
17
438
So we now know that Mira only informed SamA this morning, the same morning we learn OpenAI is to become a B-Corp and abandon its mission, and they're going to lose too more top people, Bob and Barret. Huh.
i just posted this note to openai:. Hi All–. Mira has been instrumental to OpenAI’s progress and growth the last 6.5 years; she has been a hugely significant factor in our development from an unknown research lab to an important company. When Mira informed me this morning that.
12
8
429
In all seriousness I think it is highly virtuous that they don't fix these questions even when they know we will ask them.
16
6
431
I am having difficulty reading OpenAI's 'economic blueprint' documents because the way they are written fills me with rage and makes my physically ill. I think it's important, when that kind of thing happens, to notice it.
19
8
417
You know what's weird? GPT-4o having canvas open, you tell it to make edits, it says it made edits, and there's no edits. You point out it didn't make an edit, it says it made the edit again. Still no edits. I point out it's gaslighting me. It triples down. Huh.
33
4
389
Took in all the public info. I do not know what has happened, let alone what is happening. Every time I try to put the pieces together, they don't add up unless there is something big we don't know. Does not make sense.
20
16
375
@ESYudkowsky Tell her that she'll be in Beyond the Spider-Verse and there's like a 30% chance you turn out to be right.
1
0
364
Everyone: We can't pause or regulate AI, or we'll lose to China. China: All training data must be objective, no opinions in the training data, any errors in output are the provider's responsibility, bunch of other stuff. I look forward to everyone's opinions not changing.
Just read the draft Generative AI guidelines that China dropped last week. If anything like this ends up becoming law, the US argument that we should tiptoe around regulation 'cos China will beat us will officially become hogwash. Here are some things that stood out. 🧵.
36
38
361
So let me get this straight. She is Black Widow and sued Disney, so you thought after she turned you down you'd mimic her voice without her and that would go how exactly?.
29
26
352
Only played it in a constructed event so far, but man is it fun and powerful, and deadline is coming soon. Sideboard should have more Wicked Wolves, probably.
15
120
339
I've historically said the world's greatest charity is Amazon, but on reflection I mostly do buy the argument that the right answer is instead Google.
Google has probably produced more consumer surplus than any company ever. I don't understand how a free product that has several competitors which are near costless to switch to could be the focus of an antitrust case.
10
15
337
I don't get why almost no one chooses the basic economy option on flights. Are people just vaguely afraid of it?.
100
1
330
This was worth every penny and I am worried they then didn't hire the firm going forward.
@nearcyan When I was at OpenAI we hired a firm to help us name GPT-4. The best name we got was. GPT-4 because of the built-in name recognition. I kid you not.
4
8
332
A first attempt to communicate to regular people. There's something great about doing real editing and cutting something down a word at a time to within an inch of its life. Very happy with how my first newspaper piece turned out.
27
26
303
In terms of AI actually taking our jobs, @MTabarrok reiterates his claim that comparative advantage will ensure human labor continues to have value, no matter how advanced and efficient AI might get, because there will be a limited supply of GPUs, datacenters and megawatts, and.
14
18
308
This applies to everything and is the secret. Figuring out how a given thing actually work instead of vibing - whether or not you do it because you were bad at vibing - does radically worse up front, until it suddenly starts doing radically better.
has anyone written about the phenomenon where autists go from really bad socializing to being way better than the average normie once they realize the game has rules that you can debug & create new rules?.
7
11
298
How fake? Entirely fake, it turns out. This was a test of the emergency bullshit detector system. This was only a test. How'd you do?.
That story about the AI drone 'killing' its imaginary human operator? The original source being quoted says he 'mis-spoke' and it was a hypothetical thought experiment never actually run by the US Air Force, according to the Royal Aeronautical Society.
33
23
270
This is the head of alignment work at OpenAI, in response to the question 'how do we know RLHF won't scale?'.
@alexeyguzey @SimonLermenAI We'll have some evidence to share soon.
9
16
281
I am dying at the suggestion that we could simply not teach the AI to lie, like in that movie where lying wasn't invented.
AI models will get extremely good at deceiving humans if we teach them to lie, which is what WokeAI is doing. “Trust & Safety” should be replaced with Truth & Safety.
20
4
283
I did a tally. Top answers in order:.1. First mover advantage .2. Rate limits.3. Anthropic isn't trying.4. Kleenex effect (ChatGPT names category).5. Mindshare (circular?).6. Isn't 10x better.7. Claude is a bad name.8. Lack of marketing.9. Lack of other features.10. Long tail.
Why do we think Claude is having such an epic fail at reaching consumers? Is it fixable?.
26
4
270
Alternatively, there's *definitely* no way THIS is a thing, right? Right?
15
12
249
Everyone talking now about how wise Newsom was to veto SB 1047 and how we instead need to follow his path of targeting when people use AI for particular purposes?. Remember this day, for you will rue it.
21
14
258
This is a shadow of what's to come - No one is paying enough attention to which of the frictions the AIs will remove were previously load bearing.
This is an oft-overlooked part of process design, and so I’m promoting it out of the thread. Certain forms of friction are load-bearing. Sometimes it is because they embed a proof of work, sometimes they outsource a velocity limit, and sometimes they embed a desirable tripwire.
4
10
259
Ask the girl out, though.
"get your affairs in order. buy land. ask that girl out.". begging the people talking about imminent AGI to stop posting like this, it seriously is making you look insane both in that you are clearly in a state of panic and also that you think owning property will help you.
3
2
257
I interpret this as "I cannot tell which AI startups are bogus.".
Talked to some of the startups in the current YC batch. There are a huge number of AI startups, but not one I talked to seemed bogus. There are that many legit opportunities in AI.
10
4
246
I've now done full RTFB (read the bill) on California's AB 3211. Those warning about the bill were underselling how bad this is. Yes, there is all the needless stupid notifications and icons and disclaimers, but it also de facto. bans all existing LLMs?.
7
25
249
Remember, kids: The efficient market hypothesis is false.
The thing about all of the economics papers saying there will be like .01% or .3% etc. impact on GDP from AI in 10 years, is that they're just obviously false + embarrassingly so.
12
8
238
Oh, Elon must have a good reason, they said. He's smarter than you and thinking ahead, they said. Yes, well.
This MFer posts recycled memes and replies to Ian Miles Cheong's factoids with "interesting" fifty times a day and then has the audacity to call links "lazy".
11
4
240
This is the r1 reaction thread. Share your experiences, takes, benchmarks, anything you think I might otherwise miss.
44
4
238
This is The Way. Major kudos.
I *WAS* WRONG - $10K CLAIMED!. ## The Claim. Two days ago, I confidently claimed that "GPTs will NEVER solve the A::B problem". I believed that: 1. GPTs can't truly learn new problems, outside of their training set, 2. GPTs can't perform long-term reasoning, no matter how simple.
1
7
225
I am not confident on whether we should ban or not ban non-compete agreements, but I do notice it seems like a choice that should be made by Congress, not the FTC?.
35
12
229
I see reports like this and every damn time I try to use Google Research it I get a generic report on the subject I vaguely asked about that doesn't answer the question I asked. If I challenge it, it apologizes and then does it again.
holy HELL Gemini deep research is unbelievable . just pulled info from 100ish websites and compiled a report on natural gas generation in minutes . maybe my favorite ai product launch of the last… at least three business days.
11
21
201
OpenAI skipping over 'o2' has unfortunate Kabbalistic implications.
18
9
226
And yet almost all economic debates over AI make exactly this assumption - that frontier model capabilities will be, at most, what they already are.
It seems like the narrowest of narrow possible bull eyes to assume capabilities stop exactly where we are right now. Don’t know where they go, but just predict where software adoption curves of status quo technology get to in 5 or 20 years. It’s going to be a bit wild.
14
14
225
Worth noticing: In his Bloomberg interview (~6min in) Nadella seems to imply he wouldn't let Altman pursue his various side projects as a Microsoft employee.
16
7
219
I love that @tylercowen is using 'you can run this through o1 pro yourself' as the new way of calling economic arguments Obvious Nonsense.
4
8
218
The weird part no one talks about is that they're only offering Canada one state and two senators. If we do this shouldn't the provinces be states?.
Canada as the 51st US state! 🇺🇸🇺🇸🇺🇸. Free trade and open borders with Canada? How dare those Canuks take our jobs!.
25
4
211
Oh come on, today? This is getting rediculous.
Here's my conversation with Eliezer Yudkowsky (@ESYudkowsky) about the danger of AI to destroy human civilization. This was a difficult, sobering & important conversation. I will have many more with a variety of voices & perspectives in the AI community.
15
7
207
Reminder that this corner of the world has a very different outlook than most other corners. If you don't have random day-to-day AI queries what are you even doing all day?.
Who the fuck has random day-to-day AI queries?.
8
8
214
Guy who reads up on the canonical history of Stargate and thinks 'oh all's well that ends well let's try that plan.'.
canonically the stargates are sometimes used by dangerous entities to harm us but once humanity takes care of that they end up being pretty great.
11
6
211
EA offers $100k in prizes for criticism of EA, suggests forms of criticism that accept EA framework wholesale and focus on organization/plan details. Considering submitting an entry criticizing the contest's failure to want more important criticism.
14
8
203
Daron Acemoglu is economically highly skeptical of and generally against AI. It turns out this isn’t about the A, it’s about the I, as he offers remarkably related arguments against H-1B visas and high skilled human immigration. The arguments here are truly bizarre. First he.
12
6
209
Serious thanks to OpenAI for telling us about this now and then actually doing the safety and red team testing, rather than surprising us (or skipping the testing).
o3, our latest reasoning model, is a breakthrough, with a step function improvement on our hardest benchmarks. we are starting safety testing & red teaming now.
2
10
206
I would do it for free. Exactly one of us is pricing correctly.
You'd have to pay me like $20m a year to be a US Senator.
16
0
194
I do admire Meta's commitment to the bit of always doing the worst possible thing.
12
26
194
I set out to write up the AI events of the past week and., well, things escalated quickly. Long post is long. Here you go.
7
38
202
If we want the Pro Tour done right, we've got to do it ourselves.
7
6
183
I notice I am confused.
𝕏 will soon remove the like/comment counts from replies. This change is being made to create a cleaner user interface. Additionally, 𝕏 plans to remove these counts from the news feed as well in near future.
12
5
195
For those who think the worried about AI should 'short the market': Say someone thinks 50% chance of existential risk from AI, including loss of control, by 2050. Otherwise, they are bullish, because AI. Tell me exactly what portfolio you would propose. Use tickers.
25
4
196
Given I previously supported banning TikTok, seems necessary to be clear that The Restrict Act (S 686) is a no good terrible very bad bill, and in light of it I now see I was wrong to support banning TikTok. So it goes.
6
36
195
6-0 in @TwitchRivals with Cavalier Fires featuring 4 Sphinx of Foresight. Deck is super sweet, probably should find room for 2nd Kenrith and a few SB Disenchants. Enjoying this streaming thing ( See everyone tomorrow at 3!
7
18
188
It is fascinating watching people's reactions in real time and comparing them to the reactions I was anticipating. Overall it's going great, especially people actually noticing the corrigibility problem, and often I'm watching their first reactions thinking 'wait for it. '.
New Anthropic research: Alignment faking in large language models. In a series of experiments with Redwood Research, we found that Claude often pretends to have different views during training, while actually maintaining its original preferences.
13
22
188
Lot of unforced errors suggested on why Claude isn't catching on but this seems like the clearest case so far of 'the Anthropic followers of mine should have someone implement this, ideally by Monday at the latest.'.
@TheZvi It’s possible that “share a link to this chat” played a big role in ChatGPT’s viral growth. Claude doesn’t have this feature.
15
5
196
Senator Blumenthal seems to be taking AI seriously and attempting to understand, much more so than anyone else at that level. We need to do our best to reach out and help him improve his model.
The urgency around A.I.’s advancements demands serious action & substantive legislation. The future is not science fiction or fantasy. It's not even the future, it's the here & now.
7
13
186
Like this post to indicate:.1. That you are not subject to a non-disparagement clause with respect to OpenAI or any other AI company. 2. That you are not under an NDA with an AI company that would be violated if you revealed that the NDA exists.
7
3
190
I suppose I should thank OpenAI for the timing in taking their mask off - could there be any more clear statement of the need to sign SB 1047 into law?.
4
15
187
I'm all for cryonics but I notice I want a provider who doesn't give 50% black friday discounts. .
tomorrow bio has a 50% off black friday discount if you wanna get cryonics with them btw.
5
6
191
Sigh. I just literally blew basically all of my Manifold EV and profits on one massive stupid misread cause I wasn't thinking. Totally my fault but I'm pretty disgusted.
14
2
187
I want to be clear, I did not wake up and choose violence. I spent several days considering my options.
.@TheZvi woke up and chose violence:. "Sam Altman is not playing around. He wants to build new chip factories in the decidedly unsafe and unfriendly UAE. He wants to build up the world’s supply of energy so we can run those chips. What does he say these projects will cost? Oh,
5
6
183
I noticed this too anecdotally, ridership seems substantially up with congestion pricing. <Padme> So we're adding more trains, right?.
I can't remember the last time the NYC subway felt this packed. Really annoying.
4
6
187
I love this so much. And they're only charging $9. Prices are magic.
Congestion pricing coverage is peak economics. Every person is like, "I didn't used to bear the external costs of my behavior. But now that I do, I am adjusting along an infinite variety of least-cost adaptive margins that no central planner could have foreseen or designed"
5
3
189
Quick 7-1 with a loss to flood and it's on to day 2. Played the green deck with no changes. Faced Lukka Fires at 5-1 and 6-1 and once early, Umori, All That Glitters Lurrus, Reclamation, RB Aggro, Food.
7
48
180
In the interest of the speed premium I did my best to write up what we know so far and where my model is on it. WP: Substack:.
9
35
176
VCs asking you to show them proof of who is hitting on you, if the CCP isn't trying why should they even invest?.
6
5
181
Vintage Cube is so much better viewing than every other format it's ridiculous.
6
4
163