In a June 2026 post, analyst @teortaxestex accused Anthropic of overstating its '<2X' RSI progress. The claim, posted on X, cites Anthropic's compute abundance and autonomous theorem-proving as evidence that the company's RSI timeline is misleading.
Key facts
- @teortaxestex posted accusation on X in June 2026
- Claim: 'thousands of GPUs per researcher' at Anthropic
- Analyst questions '<2X' RSI metric, asks 'relative to what?'
- Models can 'autonomously prove hard theorems' per source
- Anthropic has not publicly defined the '<2X' metric
In a June 2026 post on X, analyst @teortaxestex accused Anthropic of being "full of shit" over its claims about recursive self-improvement (RSI). According to @teortaxestex, the company has "thousands of GPUs per researcher" and no longer faces bottlenecks in kernel engineering, suggesting its models can "autonomously prove hard theorems" and design experiments. The post questions the meaning of "<2X" — likely a reference to a doubling in capability or efficiency — asking "relative to what?" The analyst concludes, "This is RSI."
The accusation lands amid a broader debate about Anthropic's compute allocation and RSI timelines. [According to public reporting], Anthropic has secured significant GPU clusters, but the company has not disclosed per-researcher ratios. The "<2X" figure appears to reference an internal metric, though Anthropic has not defined it publicly. The analyst's skepticism reflects a pattern where vendor RSI claims outpace verifiable evidence.
Key Takeaways
- Analyst accuses Anthropic of overstating RSI progress, citing GPU abundance and vague '<2X' metric.
- Credibility gap highlighted.
Why the "<2X" claim matters
The "<2X" figure is central to the dispute because it implies a specific, measurable threshold for RSI progress. [According to the source], the analyst interprets it as a capability doubling, but without a baseline, the number is meaningless. This echoes earlier debates over AI capability benchmarks, where vague metrics often mask underlying compute advantages. The analyst's point is that Anthropic's compute abundance — thousands of GPUs per researcher — makes any "<2X" claim inherently suspect, as the company can brute-force progress rather than achieve genuine algorithmic gains.
The credibility gap
Anthropic has not responded to the accusation publicly. [According to the source], the company's silence on the "<2X" definition and per-researcher GPU counts leaves the claim open to interpretation. The analyst's tone — "full of shit" — signals frustration with vendor opacity, a sentiment shared by many in the AI research community. As RSI becomes a focal point for frontier labs, external scrutiny is likely to intensify.
Anthropic declined to comment on the post.
What to watch
Watch for Anthropic's next public disclosure on RSI metrics or compute allocation. If the company releases a technical report defining '<2X' or addresses the GPU-per-researcher claim, it will clarify the dispute. Also monitor any response from @teortaxestex or other analysts.
