tristanj

8 ms·

121.
by tristanj 2mo ago
Anthropic has 80%+ margins on inference. Google has 30%+ margins on compute. Both parties have discovered a literal money printer. The payback period is <2 years. At those unit economics, anyone not borrowing aggressively here to create
122.
by tristanj 2mo ago
This was never an issue. Space-based data centers in SSO will connect to the internet via Starlink's laser network. SSO data centers were never meant for low-latency applications. The intent is to perform non-time sensitive workloads,
123.
by tristanj 2mo ago
> So call me when one of them can pass the Turing test against me and then we can talk about AGI Ring ring I'm calling you right now. We blew past the Turing test goalpost over a year ago, using 2024 models. https://www
124.
by tristanj 3mo ago
Maybe, but in the official verdict they didn't take into account all affected children either; they only used an estimated 37,500 affected teen users to calculate the fine. Meta was fined for: - 37,500 instances of deceptive-practice v
125.
by tristanj 3mo ago
This analysis actually understates how massive and outsized the penalties are. The court ruling only applies to children in New Mexico under age 18. New Mexico has ~450,000 people under age 18, of which roughly 140,000 are between ages 13-1
126.
by tristanj 3mo ago
> Now there is no sign of actual AGI happening any time soon Do you genuinely hold this position, or do you not realize how far the goalposts have shifted? In 2022, prominent AI critic Gary Marcus offered to bet $100,000 that we wouldn
127.
by tristanj 3mo ago
Public models are months behind what the labs have internally, and I believe OAI's upcoming model is so advanced they consider it AGI. OAI is not going to announce anything until their next model officially releases (allegedly later th
128.
by tristanj 3mo ago
Both of those links are literally about the exact same lawsuit in the submitted article. You just Googled the topic and pasted two sources from other media outlets thinking they were different cases. Very ironic insulting everyone’s Googlin
129.
by tristanj 3mo ago
From the article: The company must also enact certain safety features such as limiting the amount of time underage users in the state spend on Facebook and Instagram, hiding by default the number of “likes” on photos for such people and di
130.
by tristanj 3mo ago
Per the court ruling [0], the specific law that Meta violated was New Mexico’s public-nuisance law, NMSA 1978 § 30-8-1. The relevant section that was violated is [1]: A public nuisance consists of knowingly creating, performing or maintain
131.
by tristanj 3mo ago
Meta is offering a 10x discount on input ($0.10 vs. $1.25/Mtok) and 20x discount on output ($0.20 vs. $4.25/Mtok) if you opt in to let them train on your data. https://developer.meta.com/ai/models/muse-sp
132.
by tristanj 3mo ago
Anthropic is profitable, running at ~$75B ARR, and has 80+% gross margin on their API products. OpenAI is not yet profitable, but is running at ~40B ARR and their revenue is growing rapidly at >200% YoY.
133.
by tristanj 3mo ago
I used to think that export controls didn't work, until I realized that China has leapfrogged the West on every technology that wasn't subject to export controls.
134.
by tristanj 3mo ago
Kimi K3 also built Mac OS 27 in a browser. https://macos27.kimi.page/
135.
by tristanj 3mo ago
Anthropic removed raw CoT from all models since Sonnet 3.7 (released February 2025), all Opus 4.X models have censored CoT, yet Chinese labs continue to distill from them, proving this restriction wasn't an insurmountable hurdle. There
136.
by tristanj 3mo ago
Meta, Microsoft, Amazon also serve their own models, though these models are not frontier models.
137.
by tristanj 3mo ago
Encrypted reasoning traces don't prevent distillation. You only need the input prompts and final output responses to distill capabilities. As Anthropic explained in their distillation report (linked above), Moonshot AI already collecte
138.
by tristanj 3mo ago
It's based on Anthropic's press release where they reveal distillation of their models by Chinese labs: https://www.anthropic.com/news/detecting-and-preventing-dist... Moonshot AI already distilled over 3.4 m
139.
by tristanj 3mo ago
Moonshot already has at least several million exchanges distilled from Claude that they obtained over the past year https://www.anthropic.com/news/detecting-and-preventing-dist... . So they have the infrastructure alr
140.
by tristanj 3mo ago
Fable was launched on June 9 (for 72 hours), then K3 launched on July 16. Timeline wise, Moonshot had over a month to post-train K3 on Fable distilled data, which is more than enough time.
141.
by tristanj 3mo ago
No, it's a flawed conclusion. Claude Fable was publicly available for 72 hours early June. Moonshot more than enough time to prepare infrastructure, gather their preferred distillation data from Fable, and complete post-training well K
142.
by tristanj 3mo ago
Your theory doesn't hold up against the data. While model identifiers like "claude-opus-4-5-20251101" appear thousands of times across GitHub and other code sources, so do "gpt-4o-2024-08-06", "gpt-4.1-mini-202
143.
by tristanj 3mo ago
1) Model distillation is the process of transferring knowledge from a large model to a smaller one. It doesn't require logits. https://en.wikipedia.org/wiki/Knowledge_distillation 2) The word "attack" is
144.
by tristanj 3mo ago
It's clearly listed in the article I linked. The number comes from SemiAnalysis, from a newer report than the one you cited.
145.
by tristanj 3mo ago
Anthropic is currently profitable, generating around $1B/quarter and ~$50B in ARR. About 75% to 85% of Anthropic's revenue comes from its usage-based API business, which has a gross margin that exceeds 80%. https://www.
146.
by tristanj 3mo ago
Update: Aaron mentioned on X that he hasn't tested it on public ChatGPT Pro, so it was indeed an unreleased/internal model. https://x.com/aaron_lou/status/2079295442760736898 The next few weeks will be e
147.
by tristanj 3mo ago
I did read the article, but it misses the core issue entirely, and it's why I shared my comment to begin with. Look at the cost-per-task benchmarks from Artificial Analysis https://artificialanalysis.ai/models?cost=cost
148.
by tristanj 3mo ago
The people who are most afraid of Chinese models are the VCs who poured into Anthropic and OpenAI at astronomically high valuations. Anthropic is valued at $1.2T and OpenAI is targeting $850B. These astronomical valuations were built on the
149.
by tristanj 3mo ago
Not necessarily, it's plausible that GPT-5.6 Sol could also find this disproof, as neither of us has enough information to make a conclusive case. Public GPT is certainly capable, it was able to reverse-engineer the counterexample into
150.
by tristanj 3mo ago
My Fable 6 theory is admittedly speculative, but your “[public GPT-5.6] Sol is able to find the same counterexample” is also a jump in conclusions. Aaron specifically says he used “an internal version of Codex”. When asked whether that mean

More ›