How LLMs Actually Generate Text
They are just gigantic probability calculators. Everyone who thinks LLMs are sentient should be forced to watch this.
Ondrej identifies Permanent Underclass as those denied Frontier Tech
A permanent thwartation. A permanent handicap, disadvantage, restraint. “This is the second warning from the Universe in the last two weeks that [humanity must] not rely on closed-source AI and that open-source AI is the answer. …most people do not realize this because they are not capable of long-term thinking.” Severe betrayals by Big Tech … Read more
Has godaddy.com always been this bad?
godaddy.com has become Ridiculous and Irrelevant …perhaps it Always Was? Today’s transcript of incompetence. Incredible Horrendous. You at 8:41, Jun 22: Customer reports Akamai Access Denied when accessing http://www.godaddy.com/ from static public IP REDACTED. Error page is https://errors.edgesuite.net/18.dcab3717 with reference IDs 18.dcab3717 and 18.dcab3717, and requests CDN/WAF review for a possible IP block or reputation … Read more
Color Printers Tracking You
That explains why you can’t print a black and white document when you are out of yellow ink. I actually won a case against the government that sent me papers dated of a certain due date and they blamed the long delay before I received it on the post being congested and slow. But I … Read more
Predicting Consequence via Abstracted “Experience” – Limited Learning Machine
Rather than merely predicting ‘what word most often comes next’, conjoined alternate “world views” are surmised to offer ‘the next levels’ in machine learning, making our tools more anthropomorphic and adept at ‘real-world’ use. Imagine that Japanese- and Kanji-trained models that have also been trained with ‘Samurai moves’ and culturally-specific videos would be ‘tuned’ to … Read more
Paranormal Aspects of Datacenters
“…so maybe there’s more to it than They’re admitting…” — seems likely, given all else we’ve learned lately. Interesting observation from the guest: “You’re either pro-human or anti-human, and the divide is becoming much more pronounced.”
ProjIntel – A Third Leg of AI LLM CLI Harness
query — is there an additional kind of memory that could be added to be highly impactful at increasing speed and reducing llm hits? Related to “Known project map + command index + current operational facts.” Why it would help A lot of LLM/tool time is spent repeatedly answering: “Here is the known project map. … Read more
Beyond Chat History: A Multi-Layer Memory Stack for Dockerized AI Agents
Multi-Memory Harness Technique for AI LLM CLI (codex, opencode, pi, etc) PRIMARY CLAIM — Memory retrieval reduced repeated context reconstruction and likely reduced downstream LLM reasoning/search work. Two Kinds of Mem 2 kinds of ‘long-term memory’, 2 strengths, 2 uses/reasons 1 Postgres PGVECTOR 2 WIKI How the memories are best used › posit: pgv stores … Read more
How AI Short-Circuits Human Cognition – The Essay Isn’t The Point
Quite often, the process IS the point. The large language models behind ChatGPT, Gemini, Claude are at their operational core sophisticated pattern-matching systems trained on enormous quantities of text. They can summarize, translate, and simulate reasoning with a fluency that’s easy to mistake for understanding, but they don’t understand the text they generate in the … Read more