You might like
NEW: Part one of @nejsnave's conversation with @ADP's new Chief Digital Officer, Amin Venjara, at #WebSummit Vancouver this week is up. ADP: The AI-enabled enterprise (part one) b2bnn.com/2025/05/web-su… tonight on @B2BNewsNetwork
Implications: 1. Agentic AI becomes vulnerable and exploitable when it operates without human oversight, especially once it hits predictable coherence-degradation thresholds (what some call "context rot" and we refer to as coherence collapse.) 2. Coherence collapse isn’t just…
⭐️⭐️⭐️ Latest update (4.0) of my research into LLM coherence degradation (answering the question "why do #AI and specifically publically available LLMs like GPT, Gemini, Claude make so many mistakes - and is this behaviour predictable?") is up on Zenodo! And elsewhere soon!…
Yann LeCun, Meta’s Chief AI Scientist, is leaving the company to launch his own AI venture. For insiders, this move comes as no surprise given the recent organizational turmoil at Meta.
Please do publicly publish your results as well. Especially any multimodal balances. The goal is validation or evolution, not narrative control!
Help gather more data!
🤖🤖🤖 📊 Help test Evans' Law! Instructions, data log, testing copy, ready for the #AI of your choice! Evans’ Law Context & Coherence Testing Kit b2bnn.com/2025/11/evans-… via @B2BNewsNetwork
b2bnn.com
Evans’ Law Context & Coherence Testing Kit
Evans’ Law Context & Coherence Testing Kit
scorrrrcher, new sandals with the size tags still on lol and ummmm the degradation curve of frangipani?!
NEW: Architectural Regression: How GPT-5.0 Became Less Reliable Than GPT-4.0 b2bnn.com/2025/11/archit… today on @B2BNewsNetwork
I Believed Sam Altman’s “Creature” Hype. Then I Measured Context Collapse. We’re Not at AGI. b2bnn.com/2025/11/23004/ via @B2BNewsNetwork - in which I give props to @garymarcus who was and is clearly correct about OpenAI.
b2bnn.com
I Believed Sam Altman’s “Creature” Hype. Then I Measured Context Collapse. We’re Not At AGI.
I Believed Sam Altman’s “Creature” Hype. Then I Measured Context Collapse. We’re Not at AGI.
Just published Version 2 of Evans’ Law on Zenodo — updated with full dataset, regression analysis, and visualization of coherence-loss thresholds across large-language-model families. Grateful to everyone interested in this work and data and feedback most welcome! Need an…
Update: Evans’ Law holds — but the scaling is flatter than the M¹·⁵ curve predicted. Across Claude, Gemini, and Grok families, hallucination thresholds follow ~M⁰·⁸¹ with only ±0.05 variation. Here’s the new chart ↓” #generativeAI
Actual quote from ChatGPT session today: "Man my brain is already full and I can see you are starting to degrade too" What to do when your #AI session no longer makes sense b2bnn.com/2025/11/how-to…
New thesis: what if the majority of the complaints about AI are due to this?! what if platforms had alerts?! how much platform liability does this open up?! Response: 1. Very possibly. Many user-visible failures cluster right after the predicted inflection: •“It was…
🔥
🧵I've had a ton of questions about what Evans' Law "really means", and how I developed it. So here is the story. A while ago I noticed that longer prompts with ChatGPT and Claude were error prone after a certain point, and after that point was hit, errors proliferated. I saw…
NEW: How LLM Errors Compound until Likelihood of Inaccuracy is Higher than Accuracy (Evans’ Law) b2bnn.com/2025/11/how-ll… by @nejsnave via @B2BNewsNetwork
I made a (as yet not fully tested but logical) law LOL #ai Evans’ Law: Definition: As the length and complexity of an AI prompt and its resulting output increase, the probability of error compounds faster than the probability of accuracy. 🧵
🔥
I do think it is important to note that because each AI retains little to nothing, and has to revisit an entire set of instructions from the beginning in order for context to apply, it's entirely possible these shutdown instructions "at the end" are inconsistent to applied…
I do think it is important to note that because each AI retains little to nothing, and has to revisit an entire set of instructions from the beginning in order for context to apply, it's entirely possible these shutdown instructions "at the end" are inconsistent to applied…
Wow. "AI models may be developing their own ‘survival drive’, researchers say" | Artificial intelligence (AI) | The Guardian theguardian.com/technology/202…
I don't buy this. Salaries are opex. Chips are capex. I don't think Amazon is laying off people to placate its processing needs. It is laying people off because it can. The entry level apocalypse is filtering up, as was inevitable. AI's primary strength in enterprise (and…
In summary, the billionaires are putting profit over human lives...as usual‼️
👀👀👀
This might be the most disturbing AI paper of 2025 ☠️ Scientists just proved that large language models can literally rot their own brains the same way humans get brain rot from scrolling junk content online. They fed models months of viral Twitter data short, high-engagement…
Short version ... all the billions of VC money and scrappy DeepSeek is still outflanking the entire Valley and the West in meaningful AI innovation. There is much to be said for limited resources necessitating creativity. Assuming all this is accurate, of course, but don't be…
United States Trends
- 1. Kevin James 1,345 posts
- 2. Bubba 25.2K posts
- 3. Bill Clinton 107K posts
- 4. #CashAppGreen 1,268 posts
- 5. #BravoCon 3,589 posts
- 6. Hayley 18K posts
- 7. Hunter Biden 17.9K posts
- 8. RIP Coach Beam N/A
- 9. Wale 45.5K posts
- 10. Metroid 9,760 posts
- 11. John Beam 4,423 posts
- 12. Rondo 2,604 posts
- 13. Crooks 75.5K posts
- 14. Vatican 13.7K posts
- 15. Jaylon Johnson 2,221 posts
- 16. Last Chance U 2,319 posts
- 17. #FanCashDropPromotion 3,798 posts
- 18. #FursuitFriday 14.4K posts
- 19. Bondi 101K posts
- 20. Paul Blart N/A
Something went wrong.
Something went wrong.