Hitting "regenerate" on a wrong AI answer might get you the same mistake in new words. A new paper (arXiv:2608.05643) tries something else: every attempt critiques and fixes its own work before the vote. My breakdown: linkedin.com/pulse/red-pen-b… #AIresearch

Let Claude Code Work While You Sleep [now with long prompt you can copy] linkedin.com/pulse/let-claud… #vibecode #claudecode

Hit your 5-hour limit in Claude Code? You might have a new option: /low-priority It keeps your session running after the limit, at lower priority. Responses can pause while Claude waits for spare capacity, and every token still counts against your weekly limit. #ClaudeCode #AI

The same question in two languages yields two different responses. It occurs in all studied LLMs. linkedin.com/pulse/ask-ai-sa… If you or anybody you know is using AI in any non-English language, do pass this on to them. It's important to know. #AI #LLM #Multilingual

A szombati DeepDive műhelygyakorlat betelt. Köszönöm szépen a hatalmas érdeklődést.

Jev is 6 days old: an AI model that can't write. It returns decisions in 70 to 500 ms at $0.042 per million input tokens. I tested it on my hobby project and wrote up where it breaks. Repost for the engineer you know who runs LLM calls at volume. linkedin.com/pulse/jev-200x-…

200x Faster, 400x Cheaper, and It Can't Hallucinate. I ran Jev against Lexmechanic, and it's 1,729 datasheets. One test faceplanted within the hour. linkedin.com/pulse/jev-200x-… #GenerativeAI #AIEngineering #AgenticAI #SystemOne #Jev

That little clock in Claude Code is telling you whether your next message re-reads your whole context at full price or picks up from cache at 10%. I tested this thoroughly and wrote up what breaks the cache and the habits worth keeping. linkedin.com/pulse/60-minute…

Check out my latest article: The 60-Minute Clock in Claude Code linkedin.com/pulse/60-minute… via @LinkedIn

Farewell to the 4th AI Summit Budapest. It was amazing. See you all in 2027.

Idén 4. alkalommal szervezik meg Budapesten az AI Summit-ot. Most is, ahogy az előző 3 alkalommal az a megtiszteltetés ért, hogy előadó lehetek. (Holnap 14 órától 14:40-ig leszek a színpadon a Magyar Zene Házában.)

The AI you deployed has a political position: Left-libertarian. Yes, left-libertarian, on the original Political Compass test. Different labs, different continents, same corner. The easy story is bias. But is it true? #ai #politics #bias #libertarian #genai

I am now a bit more active on eYou, you are welcome to join eyou.social/r/wszp (I post all my cat pix there)

Check out my latest article: How can text we watermarked by @AnthropicAI (or anyone) and how can that be removed? linkedin.com/pulse/how-can-t… via @LinkedIn

I just added a new post on Ko-fi! ko-fi.com/I2I01SJFO0?viewima…

Check out my latest article: Fable 5 vs Sonnet 5 in Claude Code: When to Use Which (and Why Opus 4.8 Barely Makes the List) linkedin.com/pulse/fable-5-v… via @LinkedIn

Hope to meet you at reCAP 2026 in St. Leon-Rot, Germany. If you already use @SAP's Cloud Application Programming Model (CAP), this is the place to get AI-enabled. ;) #cap #ai #agenticAI #SAP #reCAP #reCAP2026

I just added a new post on Ko-fi! ko-fi.com/I2I01SJFO0?viewima…

Claude Opus 4.8 was released with an even more powerful setting, the Ultracode. I gave it an obscure research task. The results were outstanding, but a single prompt burned through the session limit in Max plan 20×. Have you tried Ultracode? #Anthropic #ClaudeCode