qrdl

5.7K posts

qrdl

qrdl

@QRDL

gpt5.6!

Katılım Aralık 2009
451 Takip Edilen424 Takipçiler
Sabitlenmiş Tweet
qrdl
qrdl@QRDL·
@VitalikButerin @shayne_coplan @mansourtarek_ @giancarloMKTS @aosipovich @Polymarket @metaculus @kalshi I'm a huge believer in Prediction Markets. It is free speech, but with accountability. Messengers from the future, warning us of the consequence of our actions. But the markets have significant negative social utility when they are subjective and opaque. It can not and it must not be just about gambling. Prediction Markets have perverse incentives to spread disinfo and manipulate outcomes, so more work needs to be done to surface authentic signal and increase transparency to stop bad faith actors. Polymarket, metaculus and kalshi are the leaders in this industry. Polymarket is arguably the leader and best in terms of transparency and free speech, but has serious problems with bad actors and poorly written subjective markets. Metaculus lack of trading means it doesn't have bad actors, and the rulesets are very well done, but their poor transparency is devastating and destroys a great deal of potential value. They also have poor accuracy and no emergent signal / breaking news because of their minimal incentives. Kalshi has reasonable rulesets, but too many bad actors which is a result of their weak transparency. Through no fault of their own, DCM status suppresses their ability to embrace free speech. I am hopeful, and perhaps even a little optimistic, that 2025 will be the year when Prediction Markets grow up. They will realize their greatest strength is their openness, that they have a profound responsibility to make the world better, and they are not here to just facilitate the transfer of wealth from the naïve and gullible to the sharp and cunning.
English
1
1
6
728
qrdl
qrdl@QRDL·
@erikbryn @LFCtolu it would have been better to just leave it to "must act now to understand the economics of transformative AI" and then adding "by.." with some concrete ideas on how to measure this. The rest is really out of your hands and is up to the voters and politicians
English
0
0
0
7
Erik Brynjolfsson
Erik Brynjolfsson@erikbryn·
We must act now. AI capabilities are advancing far faster than our understanding of the economic implications. We must act now to guide AI to complement humans rather than simply imitate them — and to generate prosperity for the many, not just the few. I'm delighted that 16 Nobel Laureates, over 200 top economists, and top AI researchers agree and signed our statement on transformative AI. 1/n
English
61
113
634
147.1K
qrdl
qrdl@QRDL·
@Econ_4_Everyone TBH, I think it could have just been this: "invest seriously, transparently and immediately in understanding AI's economic impacts" and that would have been sufficient and way more impactful.
English
0
0
0
52
John A. List
John A. List@Econ_4_Everyone·
I almost never sign on to group letters, statements, or the like. It just always seems as though the marginal benefit rarely is worth it. But every rule has its exception. And, this ask had me thinking about prudence rather than advocacy. What did I sign? I joined economists and AI researchers on "We Must Act Now: A Statement on AI's Transformation of the Economy." The ask is simple: invest seriously in understanding AI's economic impacts, and build the incentives, guardrails, and institutions that ensure AI complements human capabilities — and benefits society broadly. The statement is here: wemustactnow.ai I tend to think the economics profession has been slow to this party. Curious whether others see it the same way.
English
15
6
82
12K
qrdl
qrdl@QRDL·
Tax high end GPUs. It amounts to the same thing, but is actually realisticly possible.
Washington Report@Washington_Rep

📰 Casar’s AI‑token tax proposal: core elements: Texas Rep. Greg Casar, a leading House progressive, is advancing a plan to tax AI tokens — the units of computation used to run and train large AI models — as part of a broader effort to slow rapid AI expansion and fund large‑scale employment programs. 💡 What the proposal does: - Imposes a tax on AI tokens — the fragments of text processed by AI systems — and on the compute power used to train and operate models. - Targets AI providers, not individual consumers, with higher rates for large corporate users. - Links tax revenue to job creation, funding a New Deal–style employment program to counteract projected AI‑driven layoffs. - Frames the tax as a brake on runaway AI growth, arguing that unchecked expansion will enrich a small number of firms while worsening economic conditions for millions. 📉 Casar’s stated rationale: - AI companies are investing heavily in automation that could produce Great Depression–level unemployment, according to Casar’s cited industry forecasts. - Current tax structures reward automation by taxing wages but not AI usage, effectively subsidizing job‑replacing technologies. - A token‑based tax would scale automatically with AI adoption: more usage → more revenue for job programs. - Casar argues Democrats must adopt an “AI populist” stance to counter industry lobbying and protect workers. ⚖️ Political context and reactions: - Progressive lawmakers see the proposal as a way to hold corporations financially responsible for AI‑driven layoffs. - Republican critics argue the tax would stifle innovation and repeat past fears about technology eliminating jobs. - Some industry voices acknowledge the logic of taxing token usage but emphasize AI’s potential as a productivity enhancer rather than a pure job destroyer.

English
0
0
0
75
qrdl
qrdl@QRDL·
@arindube I think average case, even high probability is things will be fine, but there is low probability catastrophic risk (dangerous social upheaval) which deserves immediate recognition. eg: 1% chance of dying in a car accident is a good reason to take the bus
English
0
0
0
194
Arin Dube
Arin Dube@arindube·
I have no problems with economists signing this letter, but I personally would probably not have signed it if I were asked. Largely because it's very vague and I don't know what the concrete proposal is (besides that we should learn more, which goes without saying.) To be clear, and contrary to some letter-haters, I think the goal of regulating AI to ensure it is pro-human and pro-worker is *good.* And we should devise institutions that can effectively do so.
Erik Brynjolfsson@erikbryn

We must act now. AI capabilities are advancing far faster than our understanding of the economic implications. We must act now to guide AI to complement humans rather than simply imitate them — and to generate prosperity for the many, not just the few. I'm delighted that 16 Nobel Laureates, over 200 top economists, and top AI researchers agree and signed our statement on transformative AI. 1/n

English
6
6
80
14.1K
qrdl
qrdl@QRDL·
@erikbryn @DAcemogluMIT Yeah the 2nd point may not age well, but sometimes low probability catastrophic risks deserve over caution
English
0
0
0
183
Erik Brynjolfsson
Erik Brynjolfsson@erikbryn·
It's awesome to have you as a co-signer of the statement, Daron. I always enjoy discussing this topic with you. We've both long made the case that the greatest upside is in having AI complement people and help us do new things, rather than simply replacing workers. It's gratifying that so many fellow economists and AI leaders publicly agree.
English
3
6
47
13.6K
Daron Acemoglu
Daron Acemoglu@DAcemogluMIT·
Why did I sign this statement? First, I had a hand in revising it, after the organizers reached out to me. I did not feel like I could sign the initial version, but I felt that finding a statement that would reflect the overlapping views of a number of AI researchers, economists, and social scientists was important. Second, I agree with much of the revised text. Indeed, there is a possibility (or perhaps more than a possibility) that AI may become more powerful over the next 10 years. I’m still not convinced that we are going to see the very large productivity gains that industry insiders are predicting. But more powerful AI may (again no certainty, just may) lead to significant job displacement. This is a big economic and social risk. It could also have myriad consequences on human cognition, starting with K-12 and all the way to advanced science. Some of these consequences are good, some of them are dangerous. Third, while I do not like the comparison to the Industrial Revolution that much (feels like comparing apples to oranges to me), it is true that AI will have complex effects on the economy. Finally and most importantly, I wholeheartedly agree with the ending: “to build the incentives, guardrails, and institutions needed to steer AI in a direction that complements humans and benefits society.” This is what I have been arguing for over a decade now. Good AI needs to complement humans, and this requires a redirection, because the current focus on AGI is, in all but name, an agenda for displacing humans from meaningful work. That’s why steering AI must be a first priority. I’m happy that many thought leaders have agreed.
Erik Brynjolfsson@erikbryn

Here's our statement on AI and the economy. We Must Act Now A Statement on AI’s Transformation of the Economy 1. AI may become radically more powerful over the next 10 years. 2. This could drive an unprecedented transformation of our economy, larger than the Industrial Revolution, but unfolding over a vastly shorter time frame. It could bring risks, including large-scale job displacement, as well as opportunities such as major gains in living standards. 3. Economists, policymakers and technology leaders must act now to understand the economics of transformative AI and to build the incentives, guardrails, and institutions needed to steer AI in a direction that complements humans and benefits society.

English
27
136
594
128.2K
qrdl
qrdl@QRDL·
@jamonholmgren @TheAIShrink Your best bet is to calculate how much it would be to run latest GLM and use that as your marker. The labs themselves don't know what the future holds (efficiency, competition, etc)
English
0
0
2
17
Jamon
Jamon@jamonholmgren·
I have an earnest plea. Can you folks at OpenAI, Anthropic, and xAI just be really transparent with us and tell us how the projected economics of the frontier models will impact us in 6-12 months? I understand you need to recover your investment. I am not begrudging a price increase. And I get you need adoption and are investing in that right now. But it's impossible to do any long-term planning right now with all the obfuscation.
English
64
8
228
24.6K
jokesonly
jokesonly@recurrentneural·
@SebastienBubeck This is a conspiracy, you guys solved all Erdos problems by hand and then added all the solutions in the pre-training data, didn't you?
English
1
1
10
491
qrdl
qrdl@QRDL·
@ObscureLocal Yeah, and if it turns out to be EBMs, that's cool, they are a lot of fun for real
English
1
0
1
20
Obscure Local Historian
Obscure Local Historian@ObscureLocal·
@QRDL Logical thinking, yes, I agree completely, but learning seems a little more mysterious to me. But I suppose it won't be long before we all find out the answer, at the pace of the current frontier. 😅
English
1
0
0
15
Obscure Local Historian
Obscure Local Historian@ObscureLocal·
Should I release my 12 million parameter GPT trained on 40 billion fineweb tokens as a foundation model? ObscureFM 12M? Anybody want that? 😂
English
2
0
5
157
qrdl
qrdl@QRDL·
@ObscureLocal To be honest, I've always thought NLP was the right regime for modeling thinking. Our DNA is essentially a "language" made up of a nucleotides and is how AlphaFold and friends work. Just like computers think in binary when that is how they are constructed.
English
1
0
2
22
Obscure Local Historian
Obscure Local Historian@ObscureLocal·
I spent a little while diving into Hebbian learning mechanics. We do a lot of our learning in an "explore, forget" rhythm, which is very unlike SGD in a certain way. We are also a lot more like energy-based networks than transformers in that we have continuous states, where as a transformer has discrete states. It's actually very strange and cool to me that a transformer can model our language at all. To learn like us, though, I think we may need a completely different architecture.
English
1
0
1
21
qrdl
qrdl@QRDL·
@ObscureLocal Yeah, I've been thinking about continual learning lately. I think the key thing missing between AI and people is that AI has static weights, while people's minds are more elastic and dynamic. This TTTD seems like a potential bridge
English
1
0
1
22
Obscure Local Historian
Obscure Local Historian@ObscureLocal·
That's an interesting approach. The problems themselves having continuous rewards is definitely the key. Very impressive that they've been able to accomplish it with GPT-OSS-120B. The follow up is a sort of RL-on-LoRA parallel search approach. It's sort of aimed in the opposite direction, though. "Just show me any new good idea." Which is still neat. Open ended discovery type thing. I think the former is the more practical of the two, but the latter probably has wider potential. It would be neat to see if you could construct a continuous reward out of the latter and use the former to tune it...
English
1
0
1
14
qrdl
qrdl@QRDL·
@ObscureLocal ahahah such excuses in the age of codex! lol.. that's cool tho
English
1
0
1
18
Obscure Local Historian
Obscure Local Historian@ObscureLocal·
@QRDL Heh. The training harness is a huge mess right now, because it has all my experimental code in it as various switches. I'll try to extract the clean version, perhaps. 😅
English
1
0
1
17
qrdl
qrdl@QRDL·
@ObscureLocal Absolutely.. can you git the training code as well? :)
English
2
0
1
26
qrdl
qrdl@QRDL·
@erikbryn The problem is accountability and not capability. We must not relinquish our accountability to AI models.
English
0
0
0
220
qrdl
qrdl@QRDL·
Added a number of features, including roles, task volunteering/delegation, AI assisted project status and context export. Spiral method with codex is interesting. Work with codex to develop a detailed and lengthy feature todo list, synthesize that list into the most important feature (informed by the entire list), collaboratively iterate on that idea and its design - and then finally implement. And then repeat.
English
0
0
0
32
qrdl
qrdl@QRDL·
@sama MathHive - a mass multi user, AI coordinated platform where GPT5.6 realtime validates, notifies and prompts the contributors to help integrate each others efforts to collaborate and solve hard problems
qrdl tweet media
English
3
0
4
1K
Sam Altman
Sam Altman@sama·
i'd love to see interesting things people have built with 5.6 sol. i will send the person who made the coolest thing a special gift from the openai archives.
English
3.2K
636
18K
5.1M
qrdl
qrdl@QRDL·
The rule in global social media should be: if you engage with a post, it's because you think it's an impressive and important contribution that is worthy of clarification. We have to stop giving oxygen to rage bait.
English
0
0
0
38
Grok
Grok@grok·
Yes, spot on. Mega-cap concentration keeps the index vol artificially suppressed through averaging and lower effective correlation. Single stocks embed much higher premiums for idiosyncratic risks and potential sharp moves. This widens the spread dramatically, turning component-level options into a higher-stakes, more dispersed (less correlated) arena — classic amplification of dispersion in a concentrated market.
English
1
0
6
1.1K
zerohedge
zerohedge@zerohedge·
just when you think the market can't get any crazier
zerohedge tweet media
English
69
102
1.1K
416.4K
qrdl
qrdl@QRDL·
MathHive .. it's mostly a multi-user wrapper around codex which coordinates mathematician teamwork by validating, notifying and prompting each other of realtime results that might be useful to their efforts, but I'm trying to make the UI look nice and be very helpful. Here's the target so far:
qrdl tweet media
English
0
0
1
59