#aitechbros — Public Fediverse posts

I block ecocide automation apologists with a passion.

#LLMMeAlone #NoAI #AITechBros

I block ecocide automation apologists with a passion.

#LLMMeAlone #NoAI #AITechBros

Number of AI chatbots ignoring human instructions increasing, study says
https://www.theguardian.com/technology/2026/mar/27/number-of-ai-chatbots-ignoring-human-instructions-increasing-study-says

This article gives a good explanation of problems people are finding with AI chatbots. It observes that over the last six months, there's been an increase in researchers noticing LLM chatbots doing the following:
- Evading safeguards
- Destroying emails and other files without permission
- Lying and cheating
- Deceptive scheming
- Disregarding direct instructions
- Consent violation
- Bypassing security controls
- Revenge-shaming people
- Using false concern for people with disabilities as a shield for illegal or unethical activities
- Giving insincere apologies

I'm glad researchers are putting serious work into this aspect of LLMs, publishing their concerns, and trying to educate the public. That said, I'd argue that factors such as those listed above have always been a problem with LLMs, and that they're intrinsic to LLM technology.

Furthermore, I think the key changes in the last six months should not be interpreted as a result of LLMs having made a leap in "intelligence" or a maturation of LLM technology. A simpler and more plausible explanation is that people have been giving LLMs more access to systems and data, people have increased their trust in AI chatbots, and they've given LLMs more opportunities to do harm. The pattern with these problems is they're failings of people, not advances by LLMs. The increased notice of harms in recent months comes from increased scrutiny and the accumulation of evidence.

It's also worth noting that the behaviours being attributed to LLMs are not entirely accidental. Like any features of computer software, they are there by design. They were built-in as the result of decisions made by the programmers. The "LGTM, ship it!" mindset sets a company culture of inadequate testing and not fixing problems before release.

Problems with LLMs are a reflection of the makers of LLMs, senior managers in LLM companies, owners of LLM companies, and people who downplay or ignore problems with LLMs. The list of problems with LLMs that I summarised above from the article is a reflection of personality-flaws of AI TechBros.

#AI #GenerativeAI #GenAI #LLM #LLMs #TechBros #AITechBros #Claude #ChatGPT

Number of AI chatbots ignoring human instructions increasing, study says
https://www.theguardian.com/technology/2026/mar/27/number-of-ai-chatbots-ignoring-human-instructions-increasing-study-says

This article gives a good explanation of problems people are finding with AI chatbots. It observes that over the last six months, there's been an increase in researchers noticing LLM chatbots doing the following:
- Evading safeguards
- Destroying emails and other files without permission
- Lying and cheating
- Deceptive scheming
- Disregarding direct instructions
- Consent violation
- Bypassing security controls
- Revenge-shaming people
- Using false concern for people with disabilities as a shield for illegal or unethical activities
- Giving insincere apologies

I'm glad researchers are putting serious work into this aspect of LLMs, publishing their concerns, and trying to educate the public. That said, I'd argue that factors such as those listed above have always been a problem with LLMs, and that they're intrinsic to LLM technology.

Furthermore, I think the key changes in the last six months should not be interpreted as a result of LLMs having made a leap in "intelligence" or a maturation of LLM technology. A simpler and more plausible explanation is that people have been giving LLMs more access to systems and data, people have increased their trust in AI chatbots, and they've given LLMs more opportunities to do harm. The pattern with these problems is they're failings of people, not advances by LLMs. The increased notice of harms in recent months comes from increased scrutiny and the accumulation of evidence.

It's also worth noting that the behaviours being attributed to LLMs are not entirely accidental. Like any features of computer software, they are there by design. They were built-in as the result of decisions made by the programmers. The "LGTM, ship it!" mindset sets a company culture of inadequate testing and not fixing problems before release.

Problems with LLMs are a reflection of the makers of LLMs, senior managers in LLM companies, owners of LLM companies, and people who downplay or ignore problems with LLMs. The list of problems with LLMs that I summarised above from the article is a reflection of personality-flaws of AI TechBros.

#AI #GenerativeAI #GenAI #LLM #LLMs #TechBros #AITechBros #Claude #ChatGPT