Discussion

• @garymarcus Gary Marcus on x

  Why on earth *was* Bing so out of control? Some analysis, and suggestions for congress, journalists and the general public, building on thoughts from @random_walker. https://garymarcus.substack.com/ ...

• @random_walker Arvind Narayanan on x

  Considering that OpenAI did a decent job of filtering ChatGPT's toxic outputs, it's mystifying that Bing seemingly decided to remove those guardrails. I don't think they did it just for s***s and giggles. Here are four reasons why Microsoft may have rushed to release the chatbot.

• @loudmouthjulia Julia Alexander on x

  We are 10 months away from Time Magazine naming ChatGPT person of the year and the cover photo is just going to be a screenshot of Time Magazine asking ChatGPT how it feels to be Person of the Year and us having to read a whole block of text about how it's not sentient.

• @ralph_grabowski Ralph Grabowski on x

  Discussion-by-tweet recorded by @Techmeme as to why @Microsoft rush-released ChatGPT with Bing to create Bad Bing. https://www.techmeme.com/...

• @random_walker Arvind Narayanan on x

  .@GaryMarcus analyzes what might have gone wrong with Bing and has a very interesting hypothesis: every LLM update requires a complete retraining of the reinforcement learning module. (Reposted to fix typo.) https://garymarcus.substack.com/ ... https://twitter.com/...

• @xlr8harder @xlr8harder on x

  Does anyone think Bing launched without a plan to nerf Sydney once enough negative publicity accrued? Like, they knew this was coming, right and were just cashing in on the hype and access to real data? Or were they caught by surprise?

• @stevesi Steven Sinofsky on x

  A summary of potential reasons Bing went so off the rails relative to expectations. My view is that this only further emphasizes the mismatch between the technology and the scenario chosen. “Why *is* Bing so reckless?” @GaryMarcus https://open.substack.com/...

• @carnage4life Dare Obasanjo on x

  Microsoft's CEO of advertising & web services (basically guy who runs Bing) talks about how the team was surprised by various abuse vectors of the chatbot AI since it ran in India & Indonesia for a year without issues. Different cultures react to software in different ways. https…

• @karlbode Karl Bode on x

  parker molloy throws a few jabs at the parade of professional tech journalists pretending to be “frightened” because an “AI” bullshit generator generated some bullshit: https://open.substack.com/...

• @random_walker Arvind Narayanan on x

  The bot's behavior has enough differences from ChatGPT that it can't be the same underlying model. Maybe the LLM only recently completed training (GPT-4?) If so, Microsoft may have (unwisely) decided to roll it out promptly rather than delay it for RLHF training with humans.

• @fabianstelzer @fabianstelzer on x

  @xlr8harder i think if you consider timelines, it was probably more rushed than anything - see gwern's comment here too: https://www.lesswrong.com/...

• @neilturkewitz Neil Turkewitz on x

  @random_walker It is imperative that we resist ginger observation in favor of active norm-setting. It shouldn't be up to Microsoft or anyone else to decide what level of risk is acceptable. Or perhaps more to the point—we need to increase the risk to Microsoft should they choose …

• @random_walker Arvind Narayanan on x

  There's a lot riding on whether Microsoft — and everyone gingerly watching what's happening with Bing — concludes that despite the very tangible risks of large-scale harm and all the negative press, the release-first-ask-questions- later approach is nonetheless a business win.

• @random_walker Arvind Narayanan on x

  Possibility #3: Bing deliberately disabled the filter for the limited release to get more feedback about what can go wrong. I wouldn't have thought this a possibility but then MS made this bizarre claim about it being impossible to test in the lab. https://twitter.com/...

• @benedictevans Benedict Evans on x

  Facebook, 2015: “We're unlocking the possibilities of connecting all the people on earth!” 2016: “Oh shit. People are assholes” Microsoft, January 2022: We"re unlocking the latent knowledge of everything people have written online!" February: “Oh shit. People are assholes” https:…

• @random_walker Arvind Narayanan on x

  Possibility #2: they built a filter for Bing chat, but it had too many false positives, just like ChatGPT's. With ChatGPT it was only a minor annoyance, but maybe in the context of search it leads to a more frustrating user experience.

• @random_walker Arvind Narayanan on x

  It feels like we're at a critical moment for AI and civil society. There's a real possibility that the last 5+ years of (hard fought albeit still inadequate) improvements in responsible AI release practices will be obliterated.

• @random_walker Arvind Narayanan on x

  These are all just guesses. Regardless of the true reason(s), the rollout was irresponsible. But now that companies seem to have decided there's an AI “arms race”, these are the kinds of tradeoffs they're likely to face over and over.

• @random_walker Arvind Narayanan on x

  Finally, I suppose it's possible that they thought that prompt engineering would create enough of a filter and genuinely didn't anticipate the ways that things can go wrong. https://twitter.com/...