Agree, I recently updated our office's little AI server to use Qwen 3.5 instead of Qwen 3 and the capability has considerably increased, even though the new model has fewer parameters (32b => 27b)
Yesterday I spent some time investigating it:
- Gated DeltaNet (invented in 2024 I think) in Qwen3.5 saves memory for the KV kache so we can afford larger quants
- larger quants => more accurate
- I updated the inference engine to have TurboQuant's KV rotations (2026) => 8-bit KV cache is more accurate
Before, Qwen3 on this humble infra could not properly function in OpenCode at all (wrong tool calls, generally dumb, small context), now Qwen 3.5 can solve 90% problems I throw at it.
All that thanks to algorithmic/architectural innovations while actually decreasing the parameter count.
>I read up to here, but I wasn't convinced that this is the revelation that the author claims
The rest of the arguments is as weak:
1) both released open-source software
2) both don't like spam
3) both like using pseudonyms online
4) both love freedom
5) both are anti-copyright
etc.
Basically, the author found that Adam Back used the same words on X as Satoshi did in some emails (including such rare words as "dang," "backup," and "abandonware") and then decided to find every possible "link" they could to build the case, even if most of the links are along the lines of "Both are humans! Coincidence? I think not."
It's weird they spent so much time on the written word similarities, when the biggest reveal here is that Back disappears off the email lists (on a topic he is VERY interested in and has historically corresponded on) when Nakamoto appears, and then comes back when Nakamoto disappears.
Only if those similarities are indicating more than 'generic internet hacker' for both of them. You only need 23 bits to identify a person but those are 23 uncorrelated bits, and all the 'similarities' presented here are extremely strongly correlated with themselves.
The interests and writing style differentiate Mr. (Dr.?) Back from the general public, sure. But from what I’m reading, they don’t do a great job of distinguishing between 90s hackers.
“Get this, his PhD thesis dealt with a computer language called C++, just like Bitcoin papers used” seems both confused and impossibly lazy to me.
> “Scrap patents and copyright,” Mr. Back wrote in September 1997.
> Satoshi did a similar thing. He released the Bitcoin software under M.I.T.’s open-source license
Really?
Like saying “get this, his college-aged musical interests included the Urban American musical style known as ‘Hip Hop’; therefore Tupac didn’t really die and this is him.” Heavy on insinuation, light on seriousness. Strong “…you’re not from around here, are you?” vibes.
What does this kind of journalism hope to accomplish, anyway? Beyond bothering middle-aged nerds for gossip? And providing a frame for the author’s cute little sleuth jape?
“Good reason to look closer” assumes there’s good reason to pick through ancient rubble in the first place.
Similarities in style and word were common enough in small circles such as the cyphyrpunks that spawned those discussions.
Then there's not altogether unlikely chance that Satoshi is a nodding homage to Nicolas Bourbaki, each contributor holding part of a multiparty voting key.
Found his record in Russia's official company registry. This is what he officially does as an entepreneur:
56.10 — Restaurant activities and food delivery services
47.23 — Retail sale of fish, crustaceans, and mollusks in specialized stores
47.25.12 — Retail sale of beer in specialized stores
47.25.2 — Retail sale of soft drinks in specialized stores
47.29.39 — Retail sale of other food products in specialized stores, not included in other groups
68.20 — Lease and management of own or leased real estate
Money is reinvested into selling beer and fish :) Interestingly, he registered all that in 2019, just when the ransoms started.
I find it entertaining that even as part of a Russian hacking gang, the real threat is the Russian tax authorities. Regardless of how you got the money, need to pay the taxes.
Schukin isn't a very common last name (definitely not Ivanov-tier). The first name, the patronymic (his father is Maksim) and the last name all match, as well as the city (the article says he lives in Krasnodar). In fact, this Krasnodar-based entrepreneur is the only person that shows up in the search at all for "Daniil Maksimovich Schukin". Not to say the business was registered right when the ransoms started (2019). Too many coincidences if it's just a namesake.
Qwen3.5 comes in various sizes (including 27B), and judging by the posts on HN, /LocalLlama etc., it seems to be better at logic/reasoning/coding/tool calling compared to Gemma 4, while Gemma 4 is better at creative writing and world knowledge (basically nothing changed from the Qwen3 vs. Gemma3 era)
For llama-server (and possibly other similar applications) you can specify the number of GPU layers (e.g. `--n-gpu-layers`). By default this is set to run the entire model in VRAM, but you can set it to something like 64 or 32 to get it to use less VRAM. This trades speed as it will need to swap layers in and out of VRAM as it runs, but allows you to run a larger model, larger context, or additional models.
Indeed, thanks for pointing this out and the links. With the excitement I misread that it was an MR from the fork to the main project.
I don’t think I’m able to fix the title though.
I find it quite exciting to read some results in an effort to understand if TurboQuant main ideas can be applied to model weights. There are other similar projects, so we’ll see, but it seems some of this fork results look promising.
>One theory is that the knowledge required to solve the task is already stored in the parameters of the model, and only the style has to change for task success
>In particular, learning to generate longer outputs may be possible in few parameters
>we develop budget forcing to control test-time compute by forcefully terminating the model’s thinking process or lengthening it by appending “Wait” multiple times to the model’s generation when it tries to end. This can lead the model to double-check its answer, often fixing incorrect reasoning steps
Maybe, indeed, the model simply learns to insert the EOS token (or similar) later, and the capability is already in the base model
Yesterday I spent some time investigating it:
- Gated DeltaNet (invented in 2024 I think) in Qwen3.5 saves memory for the KV kache so we can afford larger quants
- larger quants => more accurate
- I updated the inference engine to have TurboQuant's KV rotations (2026) => 8-bit KV cache is more accurate
- smaller KV cache requirements => larger contexts
Before, Qwen3 on this humble infra could not properly function in OpenCode at all (wrong tool calls, generally dumb, small context), now Qwen 3.5 can solve 90% problems I throw at it.
All that thanks to algorithmic/architectural innovations while actually decreasing the parameter count.
reply