erwald

Master craftsmen paid apprentices almost next to nothing, and they were often contractually guaranteed to stick around for many years, so the teaching was a kind of wage and also a cost that could be recuperated later on. (The apprentice even often had to pay the craftsman to take them on.) None of those things are true for junior software engineers, who are paid to contribute and can leave at any moment. Also, yes apprentices often had to do chores. It is just not analogous at all.

No, 2027 was never their median forecast; it was their modal forecast. See https://blog.aifutures.org/p/clarifying-how-our-ai-timelines...

It's more like it moved from 2028-2032 to 2030-2035 (depending on the author).

I think electricity doesn't matter that much (yet) because China is bottlenecked on chips. I think the incentives/directives to build on Huawei also doesn't matter that much yet because it's still such a small percentage of compute relative to NVIDIA, even for Chinese AI companies. (But this too could matter more from 2027-2030 and on.)

Yeah, to be clear I'm pretty excited about confidential computing and startups building on it, like Tinfoil, for some use cases. I just wanted to point out it's far from adequate for some important threat models (e.g., securing model weights for data centers located abroad, I think). (It's also not super widely adopted in AI yet, but that seems to be changing, at least for inference workloads.)

Where did you get this information? I think it's wrong -- I'm pretty sure they used the Export Administration Regulations (EAR) under the Export Control Reform Act (ECRA), which is under Commerce, not ITAR which is under the State Department. See for example https://harvardlawreview.org/blog/2026/06/is-access-to-fable...

Confidential computing is not secure against a potential attacker who has physical access to the hardware. The CC security guarantees explicitly assume the attacker has no physical access.

I don't think ITAR has anything to do with any of this.

What are your general, vibes-based impressions of Mythos so far?

Seems pretty reasonable!

Thanks. I'm like 95% sure that you're wrong, and that GLM-5 was trained on NVIDIA GPUs, or at least not on Huawei Ascends.

As I wrote in another comment, I think so for a few reasons:

1. The z.ai blog post says GML-5 is compatible with Ascends for inference, without mentioning training -- it says they support "deploying GLM-5 on non-NVIDIA chips, including Huawei Ascend, Moore Threads, Cambricon, Kunlun Chip, MetaX, Enflame, and Hygon" -- many different domestic chips. Note "deploying". https://z.ai/blog/glm-5

2. The SCMP piece you linked just says: "Huawei’s Ascend chips have proven effective at training smaller models like Zhipu’s GLM-Image, but their efficacy for training the company’s flagship series of large language models, such as the next-generation GLM-5, was still to be determined, according to a person familiar with the matter."

3. You're right that z.ai trained a small image model on Ascends. They made a big fuss about it too. If they had trained GLM-5 with Ascends, they likely would've shouted it from the rooftops. https://www.theregister.com/2026/01/15/zhipu_glm_image_huawe...

4. Ascends just aren't that good

Kudos for changing your mind

Thanks. I'm like 95% sure that you're wrong (as is the parent), and that GLM-5 was trained on NVIDIA GPUs, or at least not on Huawei Ascends.

I think so for a few reasons:

1. The Reuters article does explicitly say the model is compatible with domestic chips for inference, without mentioning training. I agree that the Reuters passage is a bit confusing, but I think they mean it was developed to be compatible with Ascends (and other chips) for inference, after it had been trained.

2. The z.ai blog post says it's compatible with Ascends for inference, without mentioning training, consistent with the Reuters report https://z.ai/blog/glm-5

3. When z.ai trained a small image model on Ascends, they made a big fuss about it. If they had trained GLM-5 with Ascends, they likely would've shouted it from the rooftops.

4. Ascends just aren't that good

Also, you can definitely train a model on one chip and then support inference on other chips; the official z.ai blog post says GLM-5 supports "deploying GLM-5 on non-NVIDIA chips, including Huawei Ascend, Moore Threads, Cambricon, Kunlun Chip, MetaX, Enflame, and Hygon" -- many different domestic chips. Note "deploying".

Where did you read that it was trained on Ascends?

I've only seen information suggesting that you can run inference with Ascends, which is obviously a very different thing. The source you link also just says: "The latest model was developed using domestically manufactured chips for inference, including Huawei's flagship Ascend chip and products from leading industry players such as Moore Threads, Cambricon and Kunlunxin, according to the statement."

Where did you read that it was trained on Ascends? I've only seen information suggesting that you can run inference with Ascends, which is obviously a very different thing.

o1 mini seems to get it on the first try (I didn't vet the code, but I tested it and it works on both examples provided in the notebook, `dates` and `gabe_dates`):

    from collections import defaultdict
    
    def find_cheryls_birthday(possible_dates):
        # Parse the dates into month and day
        dates = [date.split() for date in possible_dates]
        months = [month for month, day in dates]
        days = [day for month, day in dates]
    
        # Step 1: Albert knows the month and says he doesn't know the birthday
        # and that Bernard doesn't know either. This implies the month has no unique days.
        month_counts = defaultdict(int)
        day_counts = defaultdict(int)
        for month, day in dates:
            month_counts[month] += 1
            day_counts[day] += 1
    
        # Months with all days appearing more than once
        possible_months = [month for month in month_counts if all(day_counts[day] > 1 for m, day in dates if m == month)]
        filtered_dates = [date for date in dates if date[0] in possible_months]
    
        # Step 2: Bernard knows the day and now knows the birthday
        # This means the day is unique in the filtered dates
        filtered_days = defaultdict(int)
        for month, day in filtered_dates:
            filtered_days[day] += 1
        possible_days = [day for day in filtered_days if filtered_days[day] == 1]
        filtered_dates = [date for date in filtered_dates if date[1] in possible_days]
    
        # Step 3: Albert now knows the birthday, so the month must be unique in remaining dates
        possible_months = defaultdict(int)
        for month, day in filtered_dates:
            possible_months[month] += 1
        final_dates = [date for date in filtered_dates if possible_months[date[0]] == 1]
    
        # Convert back to original format
        return ' '.join(final_dates[0]) if final_dates else "No unique solution found."
    
    # Example usage:
    possible_dates = [
        "May 15", "May 16", "May 19",
        "June 17", "June 18",
        "July 14", "July 16",
        "August 14", "August 15", "August 17"
    ]
    
    birthday = find_cheryls_birthday(possible_dates)
    print(f"Cheryl's Birthday is on {birthday}.")

For the same reason we don't want art to be 10,000x times more expensive? Cf. status quo bias etc.

Thanks, though I'll note that that article is about Blackrock encouraging/forcing its own workers to do hybrid work, not arguing that other companies should do so.

You were responding to a comment saying the world is not so coordinated by giving some examples of how coordination might happen. I gave some evidence that coordination of the type you mentioned does not seem to happen, at least for the topic being discussed, suggesting that the world is indeed not so coordinated (at least in this instance).

None of those are think tanks.

I meant evidence of them campaigning, or financing/instigating campaigns, against remote work, thereby influencing decisions of companies to implement "back to work" policies.

ETA: I agree that you did not say this was happening in your original comment, but it seems to me your comment implied that these companies were actually influencing major decisions (since that's the topic of the OP).

Do you have any evidence to back these claims up? (genuinely curious)

Can you link some think tank pieces arguing against remote work? I tried looking but couldn't find any. I found a few things but clearly none of these are part of an anti remote work effort:

an AEI interview https://www.aei.org/workforce-development/the-future-of-remo... which seems pretty balanced overall (and doesn't take a prescriptive position)

an AEI piece https://www.aei.org/research-products/report/the-trade-offs-... which seems pretty balanced too

a Heritage piece (from early in Covid) https://www.heritage.org/jobs-and-labor/report/labor-policy-... that seems mostly bullish on remote work (but mostly focuses on other issues, like labor rights)

a McKinsey report (also from fairly early in Covid) https://www.mckinsey.com/featured-insights/future-of-work/wh... which is mostly descriptive and also seems pretty balanced

a Cato piece https://www.cato.org/commentary/remote-work-here-stay-mostly... which argues in favor of remote work

Could you share the source on that?

The numbers in the article are PPP-adjusted.

Yes, that is indeed the main use case.

"Sure, you could prepare for imagined eventualities, or you could do the actual work of improving efficiency, reducing waste and unnecessary middle-men, and removing centuries old bureaucracies that are now absurdly pointless in the face of the internet. There is an underlying _desire_ for apocalypse encoded in this type of thinking."

OP was written by the person who co-founded GiveWell[1] to make charitable giving more effective, and who while running Open Philanthropy oversaw lots of grants to things like innovation policy[2], scientific research[3], and land use reform[4].

Anyway, more broadly I think you present a false dilemma. You can both prepare for tail risks and also make important marginal and efficiency improvements.

[1] https://www.givewell.org/ [2] https://www.openphilanthropy.org/focus/innovation-policy/ [3] https://www.openphilanthropy.org/focus/scientific-research/ [4] https://www.openphilanthropy.org/focus/land-use-reform/

"Grok" in AI doesn't quite describe generalization, it's more specific that that. It's more like "delayed and fairly sudden generalization" or something like that. There was some discussion of this in the comments of this post[1], which proposes calling the phenomenon "eventual recovery from overfitting" instead.

[1] https://www.lesswrong.com/posts/GpSzShaaf8po4rcmA/qapr-5-gro...

Sure it's easy -- you can use benchmarks like HumanEval, which Stability did. They just didn't compare to Codex or GPT-4. Of course such benchmarks don't capture all aspects of an LLM's capabilities, but they're a lot better than nothing!

The author of this article is Dan Hendrycks, who has a PhD in ML from UC Berkeley and was one of the contributors to the GELU activation function (among other things).

That's weird, are you looking only at the top 10 countries?

I've reproduced dwaltrib's results using World Bank data on 251 countries, and I get a Pearson's r of 0.82 and a p value of 5.6e-61 (!). I.e. a strong correlation, with high confidence. It makes sense too -- larger countries generally have more people, and more people generally generate more economic activity.

Code if you want to try yourself:

import pandas as pd

gdp = pd.read_csv("~/Downloads/API_NY.GDP.MKTP.CD_DS2_en_csv_v2_5551501.csv").set_index("Country Name")

land_area = pd.read_csv("~/Downloads/API_AG.LND.TOTL.K2_DS2_en_csv_v2_5552158.csv").set_index("Country Name")

gdp["GDP"] = gdp["2020"]

gdp["Land"] = land_area["2020"]

gdp = gdp.dropna(subset=["GDP", "Land"])

from scipy import stats

print(stats.pearsonr(gdp.Land, gdp.GDP))

#+RESULTS: : PearsonRResult(statistic=0.8151313879150333, pvalue=5.621180589722219e-61)