The performance changes are so big with the right harness that is makes sense to engineer the harness and fine-tune the model to one another from the start.

• Terminal Bench: 56.9 → 82.7 (+25.8)

• Toolathlon: 51.8 → 70.3 (+18.5)

Compared to GPT-5.6 Terra:

• Terminal Bench: Flash 82.7 vs Terra 78.4

• Toolathlon: Flash 70.3 vs Terra 53.1

• DeepSWE: Flash 54.4 vs Terra 69.6

• Agents' Last Exam: Flash 25.2 vs Terra 50.4

Trading blows with Terra, which is pretty interesting. No clear winner on these benchmarks, and wildy differeing scores. Very interesting!

Comments URL: https://news.ycombinator.com/item?id=49119559

Points: 745

# Comments: 347

Yeah, the encoder and decoder stuff is explicit, but the internal structure in generated during training. I don't think the big labs were doing this back when I did the research; no one was back in '24.

I just didn't get round to publishing for years, because I have a day job.

By the way, it still works! I tested it earlier this year on Qen3.6 and you still see improvements, so either a) no one actually paid attention, or b) it has more room to scale.

curl -X POST https://quantumlever.stream/api/magic-8-ball \ -H "Content-Type: application/json" \ -d '{"question":"Will my sampler touch the multiverse?"}'

I've wanted to hack together a "real" quantum random number generator for another upcoming project, and I got carried away a bit, and went down the 'over-engineering' cliff. So, for your nerdy enjoyment, I have documented it all up, and I added something cool for fellow "Multiple World Interpretation" followers in the Quantum Mechanics debate.

This QRNG uses sexy bits: Each is the decision of a photon to go left or right after hitting a 50:50 beam splitter. Standard kinda device, where you attenuate a light source down to single photons, offer them semi-mirror to bounce off, and see which PMT detector they hit (or which universe we ended up in ;) ). Basically, Through → bit=0. Bounce → bit=1.

As I take the MWI interpretation of Quantum Mechanics (its the more fun options), I have also built a Quantum Magic 8-Ball. Ask it a questions, and you get and receive exactly one answer here, plus every possible answer across the multiverse.

https://quantumlever.stream/oracle

Enjoy!

Comments URL: https://news.ycombinator.com/item?id=48689891

Points: 13

# Comments: 1

Author here.

I drafted it before I left for holiday, at it's not ready to publish.

It wasn't supposed to be officially posted yet, but I ran out of time before my flight.

My apologies!

I have pushed basic code to GitHub (https://github.com/dnhkng/RYS)

Some interesting areas to explore might be a combination of deleting some layers and duplicating others. i.e. reduce VRAM by dropping some layer (this works, well documented), and recovering performance by duplicating others (saves VRAM). I am not pursuing this, but it seems interesting!

The probes I used seem to help identify good configurations, but are quite noisey. A small probe set was initially used to make the scan tractable, and then the higher ranked models were retested on a set ~10x larger.

It turns out this does not help (somewhat surprisingly).

I am working with TurboDerp to integrate this into the Exllama v3 format.

I extended this to a 2×2 design (two languages × two content types) and the result is even starker: by layer 10, cross-language same-content pairs are more similar than same-language different-content pairs. The model cares about what you're saying, not what language you're saying it in.

This is also what makes layer duplication work — those mid-stack layers operate in a space where input and output distributions match, so you can loop through them without breaking anything. The encoding and decoding boundaries are where the blue walls show up in the heatmaps.

I think this says something interesting about how transformers organise computation internally. The mid-stack reasoning circuits are coherent enough that you can loop through them twice without distribution mismatch. The encoding/decoding boundaries are not.

Comments URL: https://news.ycombinator.com/item?id=47502295

Points: 1

# Comments: 0

I have Deepseek etc, but inferencing on DDR5 would take about 2-3 weeks for a simple scan. I think this works best with dense models, but it also seems ok with MoE.

@everyone: Can someone hook me up with Nvidia sponsorship?