William Breazeal

[William Breazeal @HelicalDee]

@elonmusk @minchoi For agentic coding, roughly Sol level but ~ 4.5 speed and token efficiency?

[William Breazeal @HelicalDee]

My initial impression is their foundation model pipeline in April was about a half generation behind Spud (February) which was about a half generation behind Mythos (February). So maybe 3-4 months behind OpenAI for foundation, ~ 5 months behind Anthropic.

Their mid/SFT/RL looks state of the art. Kimi 2.7 was probably key in generating high quality synthetic data.

Open question is how fast can they create a significantly larger parameter count model (2x so ~5.6T vs 2.8T, essentially a run with 4x the pre-training coherent compute).

If they can’t get a servable 6T foundation trained by October (not just a mid/SFT/RL update to the Kimi 3 foundation), the gap between Kimi and OpenAI/Anthropic is going to start opening up.

Also it looks like they are straining to serve the 3 model, 3.1 could potentially address that.

[William Breazeal @HelicalDee]

How do you know you are conscious and alive?

Can you extend that same kind of knowing to other humans? To other life forms?

Can you extend that to models?

There are two routes being explored. One is a complete understanding of physical reality. The second is through the vehicle of consciousness itself. Outward and inward paths to understanding.

[William Breazeal @HelicalDee]

@iruletheworldmo How do GPT 5.4 and Opus 4.6 do with the same prompt?

[William Breazeal @HelicalDee]

@karpathy @btaylor Huge advantage to the group that always writes as they see fit, regardless of the current tides.

[William Breazeal @HelicalDee]

@wholemars For multi-planetary success vs. the current slate of challenges, his companies are ahead in everything except for foundries (~4 to 6Q), pre/mid training chips (~8Q), robot production ramp (~6Q), and AI models (~1 to 2 Q).

[William Breazeal @HelicalDee]

@iruletheworldmo Was ~ +5 per gen. Then a wall at 130. Breaching may be due to gains in long term planning and visual reasoning syncing. There’s another potential wall around +145, but Mensa tests become statistically unreliable around 140.

tldr: wall broken, 1:100 now, test saturates at ~1:250

[William Breazeal @HelicalDee]

@tszzl Assumes

Scarcity, which is currently true

Efficient markets, which may be true enough

That the efficient allocation of scarce resources to the most productive user as signaled by purchase price is the most pro-social allocation, which may also be true enough in the aggregate

[William Breazeal @HelicalDee]

@TimJayas Eating the entire ecosystem over the next year.

[William Breazeal @HelicalDee]

@DeryaTR_ When 5.6 is available in Codex, the frontier will shift slightly.

In harness, the AA Pareto frontier for intelligence/cost is:

Codex Sol max
Codex Sol high
Codex Terra max
Grok Build 4.5 High
Codex Luna from max to low.

[William Breazeal @HelicalDee]

@DeryaTR_ What is your experience with Terra xhigh vs Sol medium?

[William Breazeal @HelicalDee]

@VictorTaelin This looks super nice. Is this one of your production prompts?

[William Breazeal @HelicalDee]

@yunta_tsai More true than ever, context quality (information density, coverage, fidelity) over quantity, especially with context compaction getting better and better.

[William Breazeal @HelicalDee]

@pmarca 5.6 Luna, OpenAI inference chips.

[William Breazeal @HelicalDee]

@elonmusk Speed, cost per token and latency.

[William Breazeal @HelicalDee]

@tszzl @wholemars That’s a very challenging. There will be extreme exponential pressures applied by the algorithms driving the winning RSI systems. Any alignment other than optimizing intelligence density, omniscience, latency, output speed, influence, reach and efficiency risks severe ablation.

[William Breazeal @HelicalDee]

@tszzl I think this has been expected for a while. A strong takeoff in a software phase, (I think likely within a year, limited by substrate saturation), then software+hardware. RSI leads (because the required domains for RSI are sufficiently verifiable) and then bootstraps broad AGI.

[William Breazeal @HelicalDee]

It’s a lot simpler.

Team A identifies people/organizations that need to be marginalized or targeted for leverage.

An associated or hired PR/propagandist creates and then packages this narrative “your team (left, right, religious, atheist…) says B is reprehensible because ‘moral rationalization’” and routes it to aligned/incentivized distribution channels.

Team identity bypasses rationality and the narrative spreads via artificially boosted vitality.

In this particular case, if you think it’s due to a lack of understanding regarding wealth creation, you’re going to get blind-sided over and over again.

[William Breazeal @HelicalDee]

@gauravisnotme Just points out the need for a missing service.

Something where you could take examples of your actual tasking and submit it confidentially for a good (enough) optimized model/harness/harness tuning recommendation.

[William Breazeal @HelicalDee]

“It is ultimately because humans provide value and democracy is a way to organize those humans so they continue to provide value.”

That’s basically:

“Humans are resources and optimal utility extraction is the guiding principle.”

“We hold these truths to be self-evident, that all men are created equal, that they are endowed by their Creator with certain unalienable Rights, that among these are Life, Liberty and the pursuit of Happiness.--That to secure these rights, Governments are instituted among Men, deriving their just powers from the consent of the governed, --That whenever any Form of Government becomes destructive of these ends, it is the Right of the People to alter or to abolish it, and to institute new Government, laying its foundation on such principles and organizing its powers in such form, as to them shall seem most likely to effect their Safety and Happiness.”

Which is closer to:

“The quality of human experience, for each human as an individual, is the paramount concern, and life, liberty and the pursuit of happiness are innate rights that cannot be traded or taken away”

The former is easier to interpret and implement, but the latter is more in line with the principles that create a culture that values democratic institutions.