It is on Google's github https://github.com/google/ax and the title comes from there.

While this one seems like an official Google project, some other projects are not, even if it is on github.com/google

A random example E.g https://github.com/google/filament#disclaimer

This is not an officially supported Google product.

Does Filament claim that it's Google's in the repo description? It's a valid title.

Fair enough, I missed that specific line. The point still stands that I wouldn't expect this to have GDM leadership backing. (If you're planning to use this at all, that matters for how much faith you should have in the product.)

it's built on top of https://github.com/agent-substrate/substrate which is not on Google github.

Agent Substrate is being moved to CNCF: https://github.com/cncf/sandbox/issues/523

Would be nice to hear where you are basing the claim 'vast majority', when different chain analysis companies have been pretty consistent in reporting illicit activity to be closer to 1% of total transaction volume [1][2]

1: https://www.trmlabs.com/fr/reports-and-whitepapers/2026-cryp...

2: https://www.chainalysis.com/blog/crypto-drug-sales-darknet-m...

From the TRM methodology:

1. excludes proceeds from crime that originate in fiat and later transfer into crypto

2. exclude transfers to blockchain addresses that have not been attributed to illicit activity

3. excludes laundering of illicit crypto proceeds

While this seems reasonable to create a metric to compare YoY, I think it's very flawed to generate an absolute figure of illicit activity (or percentage of transaction volume).

The problem doesn't exist when granting codex full access. Codex makes you choose upon first launch to "Set up default sandbox" or to "Use non-admin sandbox". The default option creates two new users and messes with NTFS permissions.

NVIDIA-Nemotron-3-Ultra-550B-A55B was released in June 4th 2026 and I think it was the largest open US model until thinking machines' Inkling (975B) was released a couple of weeks ago.

Yeah can't recommend their $19 plan, only took me a day and a half to hit the weekly usage cap. The $39 plan has 5x the limits so I recommend getting that instead. Having now tested it for a few days, it is the first of the chinese models that actually feels comparable to Opus-tier models

EU and NA models still have unlockable bootloaders.

Session was supposed to be shut down at start of July but looks like they got enough funding from donations to keep going for now.

The 30 day limit can be overridden by adding "cleanupPeriodDays": 9999 to .claude/settings.json

I've been using unsloth/gemma-4-31B-it-qat-GGUF daily for various small parsing and programming tasks using opencode and llama-server's front end. The past couple of weeks have made a big difference after google released the QAT variant and llama.cpp got support for MTP which means it is possible to now get 60-80 Tok/s with RTX 4090. The model fits in VRAM comfortably enough to keep it loaded even while browsing and having multiple programs.

110+ Tok/s as another data point on the RTX 5090 (Gemma 4 31B QAT + MTP at UD-Q4_K_XL) (at peak used 27 GB of vram)

The real lovely thing was getting 300+ Tok/s (Gemma 4 26B QAT + MTP at UD-Q4_K_XL) (at peak, I think I saw vram usage reach 21 GB of vram)

the problem of that setup is that it will run out of context pretty quick. So for coding agent it will limit your workflow very fast.

Hard to see any practical benefit to go after common crawl. The situation of freely accessible crawled data is bad enough as it is with archive.org and CC being pretty much the only available sources. We need more initiatives like them, not less. The scary thing is how the anti-AI sentiment is being used to lock things down further.

Scary how AI chokes off the very goodwill-oriented societies that it needs as its lifeblood, eh