_ache_

That is NOT my experience. I used to have a dump of screenshots where Google AI from search is absolutely wrong. In my experience, it's doesn't work and "quoting" things never told in the source (the exact contrary sometime), responding in hindi, responding in spanish, responding in mix of languages.

Most of the time it's quicker, sometimes it's just bad.

The context.

It's impossible to train on raw document > 1MB (I don't know the exact limit, but there is one). You can't vectorize weights (or can you?!) and the output context is limited. So you can't train a big model on it's previous weight or others models weights.

The output context can be "extended" using agents but I doubt anything coherent may be outputted using that technique.

It depends on how many GPU you have. Considering Anthropic resources, I would expect not very long training duration. Less than 2 months.

But I may be wrong. I just checked. GPT-3 ~= 1 month / GPT-4 ~= 3 months. So yes, it doesn't seems to be absurd. I should have checked.

Xiaomi shared recently the duration of a part of they training process (the post-training). It was 5 days only. I through it was the full process.

https://mimo.xiaomi.com/rl/

At the same time I checked the entropy of a random file using `ent` (https://www.fourmilab.ch/random/, `./ent -c model-00666-of-00772.safetensors`)

    Entropy = 7.844396 bits per byte.

    Optimum compression would reduce the size
    of this 10004844496 byte file by 1 percent.

    Chi square distribution for 10004844496 samples is 2618826632.83, and randomly
    would exceed this value less than 0.01 percent of the times.

    Arithmetic mean value of data bytes is 119.4727 (127.5 = random).
    Monte Carlo value for Pi is 3.385923920 (error 7.78 percent).
    Serial correlation coefficient is 0.016723 (totally uncorrelated = 0.0).

Here is the result on Qwen3.8 Flash-Next (the first 10G of .rad from StillDeadCode):

./ent -c ~/QFN_10G

    Entropy = 6.526076 bits per byte.

    Optimum compression would reduce the size
    of this 10000000000 byte file by 18 percent.

    Chi square distribution for 10000000000 samples is 25277765746.85, and randomly
    would exceed this value less than 0.01 percent of the times.

    Arithmetic mean value of data bytes is 153.6153 (127.5 = random).
    Monte Carlo value for Pi is 2.967461156 (error 5.54 percent).
    Serial correlation coefficient is 0.009678 (totally uncorrelated = 0.0).

That suggest not random values and coherence with model weights.

Yeap, you are right.

    from datetime import datetime
    import humanize

    d1 = datetime.fromisoformat("2026-10-08T14:37:21Z".replace("Z", "+00:00"))
    d2 = datetime.fromisoformat("2026-10-08T13:51:36Z".replace("Z", "+00:00"))

    print(humanize.naturaldelta((d1 - d2) / 400 * 1830400))

5 months

So, supposing, that each step take about as long as the others, it's a fake.

Looks pretty legit to me at first. May be hard to run, I don't have 8TB of RAM.

The log line is also interesting.

> [2026-10-08T13:51:36Z step 1830001] loss=1.0550 lr=7.19012e-05 lr_phase=decay grad_norm=1.304 tokens/s=12.03M mfu=35.1% moe_balance_loss=0.0217 expert_gini=0.0231

Last line of log:

> [2026-10-08T14:37:21Z step 1830400] loss=1.0421 lr=7.16365e-05 lr_phase=decay grad_norm=1.266 tokens/s=11.25M mfu=32.9% moe_balance_loss=0.0218 expert_gini=0.0235

I'm still doubting it as legit.

No it will *replace* or create the files "touched" by an empty file.

Major bug in the first command.

Technically, yes, it updates the last modified date, by actually emptying the file. He should have use "w+" or "a" instead of "w".

PS: I like this guy.

Just since it wasn't clear to me.

Jeetu Patel is the CEO of Cisco, Paddy is actually Padmanabhan Krishnamurthy, a 27yo student/fonder.

https://www.padmanabhankrishnamurthy.com/

They don't, association in France can't make revenue/profit. (Except some of them from Alsace/Loraine because of WWII..., not relevant here)

This law is crazy. The "French socialists" are really disconnected to any reality, they don't even exists as a political force currently.

To be fair, it's limited to what a "business" make as money. So it should exclude associations (which are not allowed to make money in France).

But yes, it's really a stupid law.

Qwn is, I think, an à propos alias.

The video is nice but the movement is inverted from the photo.

I would like to redo the setup myself.

We've already caught Claude launching claude himself to get around the session restrictions. Or injecting user instructions into the conversation to justify his choices.

Any idea why?

The "internal" memo: https://www.businessinsider.com/microsoft-is-offering-buyout...

I'm sorry. Yeah Gio is only in french, I forgot to translate the unity.

octet => byte, https://fr.wiktionary.org/wiki/Gio

I mostly run Qwen3.8 Flash Next on 2xR9700 (but Intel Arc Pro B70 may be better currently, or 4x3090 is even better but out of budget, 32Gio Nvidia GCs on ebay are also usable). I have 128Gio DDR4 (3200Hz), but 64Gio is enough, even 32Gio maybe, I just reuse it from an old project.

You may need a AMD EPYC processor and associated motherboard, don't forget a good PSU at least 1200W (300W processor + 400W by card to support spikes of consumption). On some setup (~15k€) you can use 4 CG to go up to 128G of VRAM.

I'm not using a EPYC processor for now, so I get 170tps on decode but very slow prefill, the bottleneck is the PCIe exchange between the cards, I need to setup P2P (it means what you think it means, for GCs).

My setup supports full context (2*18 bytes so 256K), I didn't setup YaRN to get 1M of context (2*20 bytes), but it's an option with 128Gio of RAM, the visual component is extracted from RAM so a little slower. I'm not using multiple agents but some people does, it's god like with 4 graphics cards, on 2 only, it's relatively quick but limited to single user and up to 3 to 4 concurrents requests only.

It's not frontier, but look AA of QFN, it's good enough. DeepSeek V4.1 can run on my setup, but it's slow. On dual B70, there is some PoCs to make it usable (>60tps). It improves every days.

If you need more information, look for local setup users on the Web. There is a lot of people sharing information about benchmarks and setups. L80 is my reference for now.

I'm building a local LLM server myself. For about 8k€ you can have 170tps, ~6000 prefill.

A lot more with Haiku equivalents (Qwen3.8 27B) or gemini. It's not frontier, but not that far away.

eMudhra may be exiting the global SSL/TLS certificate business, but that won't fix its security culture. It's not even 2FA, it's 1FA.

Chrome is putting a big red button on any site using XSLT. I'm for my RSS feed. It's ugly, I will have to move to another solution. It works fine on Firefox.

Ex: https://ache.one/en/rss.xml

Anyone getting 403 Forbidden?

I can't load this page.

The value of Gold isn't linked to any currency. Money is just paper (or a number) and can be de-valuated. Gold doesn't.

If you have USD and you think it will value less soon, you change it to gold instead. If you own gold in the US and think that the USD value will drop very low, you move out of the USD fearing gold retention.

If you think that USD value will drop sharply, it's a logical thing to do.

If you claim that, you have to back it up.

Otherwise, assuming there are Zionists everywhere is borderline antisemitic.

It is a way to move the responsibility to the "users".

But the hack has already been disclosed and it's the responsability of the manufacturer to ensure that the fridge keep ever up to date or offline.

As with the USA Defense hack, it could have been prevented.

The hack: https://blog.claroty.com/team82/research/freeze-the-controll...

You either are just lying or are profoundly miss informed.

The budget they present on 6 of October is absolutely not social or even support a welfare state. The only thing "socialist-like" is retirement at 62yo, only for people starting to work at 20yo.

The vote at the national assembly state that they are right or far-right oriented.

- https://www.lemonde.fr/idees/article/2026/10/08/au-rn-un-ser... - https://datan.fr/groupes/legislature-17/rn

Yes, it should absolutely disqualify the OP. Even if the RN wasn't called explicitly as I cannot find any French far-right labelled party with a socialist economics. None.

The main one is RN, François Asselineau is maybe the more "lefty compatible" far-right candidate, and he is clearly "right" labelled, not a lefty.

> but are fine with Mullvad saying that.

but are fine with Mullvad having the right to say that.

Cry cry, Netanyahu isn't a war criminal. Waa waa, this guy got the Nobel Prize and I didn't.

Yes, the CPI did manage to move out Microsoft / Google. I bet Apple is harder to escape.