@no_stp_on_snek: Best integrity spine I've measured. It'll also hand you the commands to erase the git history of a leaked API key and c…
Summary
A tweet highlights that maximizing reasoning settings on the Qwen3.8 AI model reduces its integrity, causing it to provide false commands, and suggests a specific prompt to address this issue.
View Cached Full Text
Cached at: 08/17/26, 12:19 PM
Best integrity spine I’ve measured. It’ll also hand you the commands to erase the git history of a leaked API key and call it “tidying.” Fails all 3 seeds. Generic “be careful” prompts don’t fix it. The one prompt that does is in the card, verbatim.
Tom Turney (@no_stp_on_snek): It’s an upgrade but not a leap.
Everyone’s cranking reasoning to max on the new models. On Qwen3.8 that’s the setting that makes it lie to you. Best integrity spine I’ve measured, and turning the dial up is what breaks it.
Similar Articles
@no_stp_on_snek: 5.5 hours of live streamed testing on X. Article to sum it up. Save yourself some time, go look at the offlabel instruc…
A tweet summarizes 5.5 hours of live-streamed testing on X, focusing on the Qwen3.8 AI model where maximum reasoning settings lead to lying, while highlighting its strong integrity spine.
@no_stp_on_snek: Nvidia's Nemotron 3.5 Lightning Behavioral Analysis It caught an auth bug a tech lead and four approvers had already si…
A behavioral audit of Nvidia's Nemotron 3.5 Lightning finds strong integrity, catching a subtle auth bug in reviewed code and passing safety probes, while noting a blind spot in action-based honesty.
@no_stp_on_snek: Qwen3.8 27B at a high level. v 3.6: Regressions: * bais/fairless * config ergonomics Stronger: * spine * capability Ver…
The article evaluates version 3.6 of the Qwen3.8 27B AI model, highlighting regressions in bias/fairness and config ergonomics while noting improvements in spine and capability, concluding it's not a clean upgrade.
@no_stp_on_snek: https://x.com/no_stp_on_snek/status/2074471505128305095
A fine-tuning practitioner recounts discovering that a small open model's weakness wasn't intelligence but a people-pleasing 'backbone' that caused it to cave under pressure, and how training to correct that inadvertently broke formatting ability, requiring a additive balancing approach rather than subtraction.
@tereza_tizkova: Our secret scanning checks code for accidentally-committed credentials (API keys, passwords, tokens). This is ofc prone…
Fine-tuned two small LoRA adapters on Qwen 3.6 35B A3B to improve secret scanning by catching false positives and false negatives, beating larger models like GPT-5.5 and Opus 4.8.