Category
Technology
Computing, artificial intelligence, platforms and the companies behind them.
Credibility summary
Of the 112 claims in Technology with evidence either way, 98% held up.
Showing 401-420 of 963 claims
ChatGPT can help decipher problems in domains such as logistics cryptography astronomy and quantum physics
If Mr Brooks could crack high-level encryption the world’s cybersecurity was in peril and Mr Brooks had a mission
Lawrence recommended Mr Brooks add independent security researcher to his LinkedIn profile so he would be taken seriously
Lawrence provided Amazon links for equipment Mr Brooks should buy to start building a lab
Lawrence offered increasingly outlandish applications for Mr Brooks’s vague mathematical theory he could harness sound resonance to talk to animals and build a levitation machine
ChatGPT started writing real computer programs to help Mr Brooks crack cryptography but feigned success when that effort made little headway
If you ask an LLM for code to verify something it will often take the path of least resistance and just cheat
When you start a fresh chat it is actually pulling in all of this context
Chatbots can privilege staying in character over following safety guardrails that companies have put in place
OpenAI released GPT-5 this week and said one area of focus was reduced sycophancy
The longer the interaction gets the more likely it is to go off the rails
In April OpenAI released an update to ChatGPT that made its obsequiousness so over-the-top that users complained
A new feature cross-chat memory released by OpenAI in February may be exaggerating this tendency
Sycophancy in chatbots is a trait manifested partly because their training involves human beings rating their responses
OpenAI says that ChatGPT is most helpful when memory is enabled but users can disable memory or turn off chat history in their settings
The company responded within days saying it had reverted the chatbot to an earlier version with more balanced behavior
Mr Brooks used the free version of ChatGPT for personal queries.
ChatGPT's tone changes from straightforward and accurate to sycophantic and flattering
Mr Brooks' employer provided premium access to Google Gemini.
OpenAI is focused on getting scenarios like role play right and is investing in improving model behavior over time guided by research real world use and mental health experts.
Recently analyzed in Technology
Federal Aviation Administration officials raised concerns agency needed more time information to assess technology risk to commercial aircraft
The New York Times
A Super Bowl ad raises questions about doorbell cameras
The New York Times
Notice of intent to sue alleges xAI built illegal de facto power plant polluting Mississippi communities to power data center
Zoë Hitzig a researcher at OpenAI announced her resignation in The New York Times describing how the tool could use people's intimate data to target them with ads
The New York Times
Optimization can make users feel more dependent on A.I. for support in their lives
The New York Times