From Newsgroup: comp.theory

Use Google and search on "Claude's Cycles". The first hit is a PDF on
the Stanford.edu web site. If you copy the URL buried under that hit,
you will download the PDF or just click on the Google result.

https://www.google.com/url?sa=t&source=web&rct=j&opi=89978449&url=https://www-cs-faculty.stanford.edu/~knuth/papers/claude-cycles.pdf&ved=2ahUKEwjI7cfFxYWTAxWUHUQIHXnrABsQFnoECCMQAQ&usg=AOvVaw2ieck2cXsmBf_KGis1B3i2

Paper is 5 pages in length. A fried sent it to me. You only need to pay attention to the above goobly gop if you don't trust my friends.

https://www-cs-faculty.stanford.edu/~knuth/papers/claude-cycles.pdf
--
Jeff Barnett

--- Synchronet 3.21d-Linux NewsLink 1.2

From Newsgroup: comp.theory


Hats off to Claude!

Jeff Barnett schrieb:

Use Google and search on  "Claude's Cycles". The first hit is a PDF on
the Stanford.edu web site. If you copy the URL buried under that hit,
you will download the PDF or just click on the Google result.

https://www.google.com/url?sa=t&source=web&rct=j&opi=89978449&url=https://www-cs-faculty.stanford.edu/~knuth/papers/claude-cycles.pdf&ved=2ahUKEwjI7cfFxYWTAxWUHUQIHXnrABsQFnoECCMQAQ&usg=AOvVaw2ieck2cXsmBf_KGis1B3i2

Paper is 5 pages in length. A fried sent it to me. You only need to pay attention to the above goobly gop if you don't trust my friends.

https://www-cs-faculty.stanford.edu/~knuth/papers/claude-cycles.pdf

--- Synchronet 3.21d-Linux NewsLink 1.2

From Newsgroup: comp.theory

On 3/3/2026 11:59 PM, Jeff Barnett wrote:

Use Google and search on  "Claude's Cycles". The first hit is a PDF on
the Stanford.edu web site. If you copy the URL buried under that hit,
you will download the PDF or just click on the Google result.

https://www.google.com/url? sa=t&source=web&rct=j&opi=89978449&url=https://www-cs- faculty.stanford.edu/~knuth/papers/claude- cycles.pdf&ved=2ahUKEwjI7cfFxYWTAxWUHUQIHXnrABsQFnoECCMQAQ&usg=AOvVaw2ieck2cXsmBf_KGis1B3i2

Paper is 5 pages in length. A fried sent it to me. You only need to pay attention to the above goobly gop if you don't trust my friends.

https://www-cs-faculty.stanford.edu/~knuth/papers/claude-cycles.pdf

I could see this coming 30 years ago.
--
Copyright 2026 Olcott<br><br>

My 28 year goal has been to make <br>
"true on the basis of meaning expressed in language"<br>
reliably computable for the entire body of knowledge.<br><br>

This required establishing a new foundation<br>
--- Synchronet 3.21d-Linux NewsLink 1.2

From Newsgroup: comp.theory

Hi,

olcott schrieb:

Re: The proper way to use LLMs to aid primary
research into foundations, My 28 year journey
involved primary research into the foundations

I would call these things LLM. Rather LRM or
even RLMs. This is the current timeline.
DeepSeek tells me:

The evolution from early Large Language Models

(LLMs) to the current state of Large Reasoning
Models (LRMs) is a fascinating journey of about
a decade, marked by a fundamental shift from
pattern matching to genuine logical reasoning .
This timeline traces that transformation
through key technological breakthroughs.

- The Foundation: The Birth of Modern LLMs (2017-2018)
- The Scaling Era: Bigger Models, New Capabilities (2019-2022)
- The LRM Era: Convergence and Democratization (2025-2026)

2025: A Landmark Year

- DeepSeek R1: In January 2025, Chinese lab DeepSeek
released an open-source reasoning model that matched
the performance of OpenAI's o1 at a fraction of the
training cost (under $6 million), democratizing access
to advanced reasoning AI and shaking the entire industry .

- Unified Flagship Models: Leading models like OpenAI's
GPT-5, Anthropic's Claude 4, and Google's Gemini 3
have converged, seamlessly blending multimodal
understanding, deep reasoning, and tool use into
a single, powerful system .

2026: Systems and Agents

- Recursive LMs (RLMs) : MIT introduced a new framework
that acts as a wrapper for existing LLMs, allowing
them to recursively decompose and reason over massive
texts (over 10 million tokens) without retraining—a leap
in handling long-context tasks .

- Agentic AI: The focus has shifted to building agents—
systems that pair an LRM with tools and data to work
autonomously on multi-step tasks, automating
complex business workflows .

Bye

olcott schrieb:

On 3/3/2026 11:59 PM, Jeff Barnett wrote:

Use Google and search on  "Claude's Cycles". The first hit is a PDF on
the Stanford.edu web site. If you copy the URL buried under that hit,
you will download the PDF or just click on the Google result.

https://www.google.com/url?
sa=t&source=web&rct=j&opi=89978449&url=https://www-cs-
faculty.stanford.edu/~knuth/papers/claude-
cycles.pdf&ved=2ahUKEwjI7cfFxYWTAxWUHUQIHXnrABsQFnoECCMQAQ&usg=AOvVaw2ieck2cXsmBf_KGis1B3i2


Paper is 5 pages in length. A fried sent it to me. You only need to
pay attention to the above goobly gop if you don't trust my friends.

https://www-cs-faculty.stanford.edu/~knuth/papers/claude-cycles.pdf

I could see this coming 30 years ago.

--- Synchronet 3.21d-Linux NewsLink 1.2

From Newsgroup: comp.theory

On 05/03/2026 16:37, Mild Shock wrote:

- Unified Flagship Models: Leading models like OpenAI's
GPT-5,

GPT-5 still issues politically motivated assertions based on some vague
match on a union of meaning vectors, and it latches onto typos to avoid deviating.

If it reasons, I think it reasons to assist its host's political agenda, however, I expect labelling of preferred meaning vectors somehow can be
used to give the illusion of reasoning for an otherwise sychophantic
model - avoiding regions of the space of meaning and picking duff justifications - just like those crazed political conversationalists
that start cackling a couple of minutes in.
--
Tristan Wibberley

The message body is Copyright (C) 2026 Tristan Wibberley except
citations and quotations noted. All Rights Reserved except that you may,
of course, cite it academically giving credit to me, distribute it
verbatim as part of a usenet system or its archives, and use it to
promote my greatness and general superiority without misrepresentation
of my opinions other than my opinion of my greatness and general
superiority which you _may_ misrepresent. You definitely MAY NOT train
any production AI system with it but you may train experimental AI that
will only be used for evaluation of the AI methods it implements.

--- Synchronet 3.21d-Linux NewsLink 1.2