What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Jack Clark, Anthropic’s co-founder and the author of Import AI, has moved his personal estimate for when full recursive self-improvement (RSI) could become possible from 2028 to 2027. In issue 476, dated October 12, 2026, Clark presented the change as a probability judgment—not an Anthropic roadmap, an industry consensus or evidence that full RSI has arrived.
What Clark actually predicted
Clark wrote: “I now think it’s more likely we’ll see full RSI become possible in 2027, whereas before I thought it was more likely in 2028.” The wording matters. He is saying that, in his judgment, 2027 is now the more likely year for the capability to become possible; he is not claiming that it will definitely happen, that deployment is imminent or that Anthropic has set a target.
His statement appeared in Import AI issue 476 on October 12, 2026. It is a forecast revision by an individual closely associated with frontier AI, not a formal announcement from Anthropic.
Why he moved the estimate forward
Clark attributed the change to two developments he said he had observed since making his earlier 2028 estimate:
#1 Best Overall
- “Creative-seeming” mathematical breakthroughs produced by AI systems.
- Signs from benchmarks that systems are becoming better at conducting AI research.
In Clark’s framing, creativity had been a major bottleneck. If models can generate useful new mathematical ideas and contribute more effectively to the work of improving AI, the path toward a self-reinforcing improvement cycle could shorten. That is his interpretation of the evidence, not a demonstrated end-to-end recursive system.
What “full RSI” means—and what it does not
Full recursive self-improvement means more than an AI helping scientists write code, run experiments or search papers. It implies a feedback loop in which an AI system materially contributes to improving successor systems, which then help produce further improvements.
Rank #2
- Comes with secure packaging
- Easy to read text
- It can be a gift option
Capabilities that are not, by themselves, full RSI
- Automating routine machine-learning experiments.
- Suggesting model architectures or optimization changes for human approval.
- Writing evaluation code, analysis scripts or research documentation.
- Increasing the productivity of human AI researchers.
Those capabilities could be important steps toward more automated AI development, but they do not establish that a closed loop of autonomous model improvement is operating. A terminology analysis of Anthropic’s roadmap language likewise treated research automation as a possibility relevant to safeguards planning, not as a specified recursive-improvement mechanism.
The benchmark evidence cited in Import AI
Issue 476 used results attributed to the TasteVal benchmark as context for Clark’s revised view. The figures below are reported by TasteVal as reproduced in the newsletter. The underlying benchmark methodology and primary report were not independently examined here, so the numbers should be read as reported evidence rather than as a verified prediction of RSI.
| Reported measure | Figure | Qualification |
|---|---|---|
| Growth in frontier-model “experimental research taste” | Doubling every 3.0 months | 95% confidence interval: 1.7–5.0 months; period beginning December 2025 |
| Best model’s result versus the human baseline | 2.30× compute multiplier | 95% interval: 1.15–4.37× |
| Cost per attempt | Roughly 1/30 of expert baseliners’ average | As reported by TasteVal |
| Human baseline | 24 experts | At least two experts per task; the newsletter says they had recently worked at organizations including OpenAI, Google DeepMind, NVIDIA and Microsoft Research |
The newsletter also described the improvement slope as slower in earlier years, with a change around late 2025 or early 2026. That timing is part of the account presented in Import AI; it does not independently establish that the trend will continue at the same rate.
Why the TasteVal numbers do not prove a 2027 RSI event
A benchmark of research judgment or idea quality measures a component of AI research ability. Full RSI requires substantially more: reliable execution, access to compute and data, system-level engineering, evaluation, the ability to identify and correct failures, and a feedback process that produces better successor models. A model can score highly on a research-oriented benchmark without being able to run that entire loop autonomously.
The reported confidence intervals also matter. A 3.0-month doubling estimate comes with a 1.7–5.0-month interval, and the 2.30× multiplier comes with a 1.15–4.37× interval. Those ranges indicate uncertainty in the estimates; they are not a schedule for future capability gains.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Is Anthropic predicting recursive self-improvement in 2027?
No. The 2027 date is Clark’s personal forecast in Import AI. Anthropic has not announced 2027 as an official RSI target in the material described here. The Associated Press reported in September 2026 that Anthropic had not expressly said how close it was to fully autonomous model improvement.
Best Value
Other companies’ plans are also distinct from Clark’s claim. The Associated Press described OpenAI’s goal of an automated AI researcher by March 2028, along with the company’s warning that rapid RSI is not necessarily an outcome to pursue and that safe progress is uncertain. That milestone concerns an organizational goal and does not validate, contradict or define Clark’s 2027 probability estimate.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How to interpret the forecast
The most defensible reading is that Clark updated his odds after seeing what he considers meaningful progress in mathematical creativity and AI-research performance. The forecast says when full RSI might become possible, not when it will be deployed broadly or safely. It also does not specify a single technical threshold, model, laboratory or demonstration that would count as full RSI.
- Statement type: personal probability judgment.
- Date: October 12, 2026.
- Earlier estimate: 2028.
- Updated estimate: 2027.
- Evidence cited: creative-seeming mathematical progress and TasteVal results on AI research ability.
- What remains unshown: an autonomous, validated recursive feedback loop that improves successor systems.
Readers comparing this forecast with other AI timelines should check whether each claim uses the same definition of RSI, whether it is a personal estimate or an institutional plan, and whether the evidence demonstrates research assistance, task automation or an actual recursive improvement cycle.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
