Google's "code red" over ChatGPT and the Bard launch that followed
The New York Times reported on 2022-12-21 that Google management had declared a "code red" over ChatGPT and that Pichai had upended the work of numerous groups, reassigning research and…
- Date
- 21 December 2022
- Who
- Google, Alphabet
- People
- Sundar Pichai, Jeff Dean
- Confidence
- Medium (internal reporting) / High (Bard dates)
- Deep dive
- RLHF and instruction tuning (how base models became assistants)
Tier: Supporting · Significance: 3/5 · Org(s): Google, Alphabet · People: Sundar Pichai, Jeff Dean · Confidence: Medium (internal reporting) / High (Bard dates) The New York Times reported on 2022-12-21 that Google management had declared a "code red" over ChatGPT and that Pichai had upended the work of numerous groups, reassigning research and trust-and-safety teams to AI prototypes (9to5Google summary). A week earlier CNBC had reported an all-hands where Pichai and Jeff Dean said Google had similar capabilities but a larger reputational downside (CNBC, 2022-12-13). Google announced Bard, powered by a lightweight version of LaMDA, to trusted testers on 2023-02-06 (Pichai) and opened it in the US and UK on 2023-03-21 (Google). The fact that matters for this chapter is that Google had instruction tuning (FLAN, B05-08), classifier-filtered dialogue (B05-12) and, at DeepMind, RLHF agents (GopherCite and Sparrow, B05-13a, B05-16), but shipped a LaMDA-based Bard first, from a different organization; the Brain and DeepMind groups merged on 2023-04-20 (B05-27a); a published Google post-training description with an SFT, reward-model and RLHF stage arrives with Gemini in December 2023 (B05-40). Sources: 9to5Google · CNBC · Google Bard post