Bara att kolla AI nyheterna en vanlig vecka för att se vilken vansinnig acceleration som är. Tanken på att det har peakat nu är skrattretande när man tittar på t.ex. Claude Coworker eller Chat GPT Codex. Om man bara chattar med dom offentliga gratisversionerna då och då kan man förlåtas för att tro att inte mycket har hänt men det sker alltså vansinniga kapacitets-hopp som du kan ta del av om du betalar 20-200 dollar i månaden för premium versionerna. Nu kommer både GPT Codex 5.2 och Opus 4.6 inom några veckor eller max tre månader skriva helt nya versioner av sig själva precis som Scott Alexander på SlateStarCodex förutspådde 2020 när dom första GPT modellerna blev allmänt gods. "2025 skriver GPT6 en ny version av sig själv sen rasar det iväg". Själv tror jag inte på Microsoft utspelet om att automatisera alla revisorer och programmerare på 18 månader men det är intressant att se hur diskussionen nu börjar närma sig mainstreamen. Än så länge är detta ett litet delforum på Flashback där vi pratar om detta, men snart kommer AI vara den största politiska/kulturella frågan framför alla andra.
Citat:
Claude Opus 4.6 Escalates Things Quickly Life comes at you increasingly fast. Two months after Claude Opus 4.5 we get a substantial upgrade in Claude Opus 4.6. The same day, we got GPT-5.3-Codex.
That used to be something we’d call remarkably fast. It’s probably the new normal, until things get even faster than that. Welcome to recursive self-improvement.
https://www.lesswrong.com/posts/5JNjHNn3DyxaGbv8B/claude-opus-4-6-escalates-things-quickly
Citat:
A few months later, the guesses from Mantic’s prediction engine and the other tournament participants were scored against the real-life outcomes and one another. The AI placed eighth out of more than 500 entrants, a new record for a bot. “It was an unexpected breakthrough” according to Toby Shevlane, Mantic’s CEO. Shevlane told me that he left a cushy gig as a research scientist at Google DeepMind to co-found the company. He wanted to celebrate the AI’s triumph, but he worried that it had been the product of some lucky guesses. He and his team entered a new version of it into the Metaculus Fall Cup. That bot did even better. Not only did it finish fourth, another record, it beat a weighted average of all human-forecaster predictions. It proved itself wiser than the wisdom of a pretty wise crowd.
https://www.theatlantic.com/technology/2026/02/ai-prediction-human-forecasters/685955/
Citat:
But LLMs are not used in a vacuum. Today, they are used as replacement *for a competent junior associate*. So, yes - I have to check GPT-5.2 Pro's work. But I would also have had to spend the same (or longer) amount of time checking a junior associate's work. And very importantly, a junior associates' work is often worse than that produced by GPT-5.2 Pro. When deciding whether to spend time checking GPT-5.2 Pro's work vs. a junior associate's work, I very often find myself these days choosing the former.
https://x.com/deredleritt3r/status/2022330823471505781
Citat:
Over the past five months, our team has been running an experiment: building and shipping an internal beta of a software product with 0 lines of manually-written code.
The product has internal daily users and external alpha testers. It ships, deploys, breaks, and gets fixed. What’s different is that every line of code—application logic, tests, CI configuration, documentation, observability, and internal tooling—has been written by Codex. We estimate that we built this in about 1/10th the time it would have taken to write the code by hand.
Humans steer. Agents execute.
https://openai.com/index/harness-engineering/
Citat:
At current rates of compute growth and algorithmic progress, this model's median prediction is >99% automation of AI R&D in late 2032. Most simulations result in a 1000x to 10,000,000x increase in AI efficiency and 300x-3000x research output by 2035. I therefore suspect that existing trends in compute growth and automation will still produce extremely powerful AI on "medium" timelines, even if the full coding automation and superhuman research taste that drive the AIFM's "fast" timelines (superintelligence by ~mid-2031) don't happen.
https://www.lesswrong.com/posts/uy6B5rEPvcwi55cBK/research-note-a-simpler-ai-timelines-model-predicts-99-ai-r