3 ms·

> Would be curious to hear from some Googlers on their thoughts. I'm sure, internally, a lot of it must feel like piling on from the outside, but in all honestl

by danans 4y ago

> Would be curious to hear from some Googlers on their thoughts. I'm sure, internally, a lot of it must feel like piling on from the outside, but in all honestly it really feels to me like a classic case of "big company that lost its way Former Googler, opinions are my own. They haven't lost their way technologically - as you mentioned they invented the Transformer - and internally Google has long had language models that rival ChatGPT in sheer size and coherence of responses (hallucinations and all). Bard is an intentionally toned down version of LamDa. The reason they didn't release their LLM earlier was likely due to the serious brand risk associated with making it part of Google search. Bing/ChatGPT had no such brand risk, and released their LLMs using the "There's no such thing as bad publicity" logic. That works great as a wrecking ball, but it's not a long term product strategy. So the real institutional problem at Google isn't lack of technological innovation, it's the inability to take major product risks, especially in anything adjacent to Search.

jatins 4y ago
Microsoft has been really smart in this regard because they are invested in OpenAI but OpenAI does not have to suffer from any Big tech organizational nonsense the way Google's AI probably has to
TapWaterBandit 4y ago
> The reason they didn't release their LLM earlier was likely due to the serious brand risk associated with making it part of Google search. Bing/ChatGPT had no such brand risk, and released their LLMs using the "There's no such thing as bad publicity" logic. That works great as a wrecking ball, but it's not a long term product strategy. Not sure if this is the right read considering that CHATGPT/Bing now constitute a far greater brand risk to Google than they would if Google they had gotten out ahead on LLMs. What may have seemed like prudent caution to protect a brand has now shown to be much closer to incumbent complacency. Suppose it is the classic story of big companies that get disrupted anywhere.
pantalaimon 4y ago
Well it seems like Bing’s version is more sassy than ChatGPT's Assistent. But that just makes it feel more human-like which is impressive. Google used to be quirky and exiting too, now they are trying to be as boring as possible.
- danans 4y ago
  > But that just makes it feel more human-like which is impressive. Sure, it's being sassy is impressive, but it's not useful (unless the utility is novelty or entertainment), which is not what a Search engine's job is.
  - int_19h 4y ago
    Thing is, the utility is entertainment for a lot of people. But Bing actually has a mood selector displayed prominently right about the input box, so you can set it to be more pedantic or more "creative".
Shaggy2000 4y ago
[dead]
nopinsight 4y ago
GPT-4 got much better results on many benchmarks than PaLM, Google's largest published model [1]. PaLM itself is probably quite a bit better than LamDa in several tasks, according to a chart and a couple of tables here: https://arxiv.org/abs/2204.02311 https://arxiv.org/abs/2204.02311 It's unclear that Google currently has an internal LLM as good as GPT-4. If they do, they are keeping quiet about it, which seems quite unlikely given the repercussion. [1] GPT-4's benchmark results vs PaLM: https://openai.com/research/gpt-4 https://openai.com/research/gpt-4

by danans 4y ago

jatins 4y ago

Microsoft has been really smart in this regard because they are invested in OpenAI but OpenAI does not have to suffer from any Big tech organizational nonsense the way Google's AI probably has to

TapWaterBandit 4y ago

> The reason they didn't release their LLM earlier was likely due to the serious brand risk associated with making it part of Google search. Bing/ChatGPT had no such brand risk, and released their LLMs using the "There's no such thing as bad publicity" logic. That works great as a wrecking ball, but it's not a long term product strategy. Not sure if this is the right read considering that CHATGPT/Bing now constitute a far greater brand risk to Google than they would if Google they had gotten out ahead on LLMs. What may have seemed like prudent caution to protect a brand has now shown to be much closer to incumbent complacency. Suppose it is the classic story of big companies that get disrupted anywhere.

pantalaimon 4y ago

Well it seems like Bing’s version is more sassy than ChatGPT's Assistent. But that just makes it feel more human-like which is impressive. Google used to be quirky and exiting too, now they are trying to be as boring as possible.

danans 4y ago
> But that just makes it feel more human-like which is impressive. Sure, it's being sassy is impressive, but it's not useful (unless the utility is novelty or entertainment), which is not what a Search engine's job is.
- int_19h 4y ago
  Thing is, the utility is entertainment for a lot of people. But Bing actually has a mood selector displayed prominently right about the input box, so you can set it to be more pedantic or more "creative".

Shaggy2000 4y ago

[dead]

nopinsight 4y ago

GPT-4 got much better results on many benchmarks than PaLM, Google's largest published model [1]. PaLM itself is probably quite a bit better than LamDa in several tasks, according to a chart and a couple of tables here: https://arxiv.org/abs/2204.02311 https://arxiv.org/abs/2204.02311 It's unclear that Google currently has an internal LLM as good as GPT-4. If they do, they are keeping quiet about it, which seems quite unlikely given the repercussion. [1] GPT-4's benchmark results vs PaLM: https://openai.com/research/gpt-4 https://openai.com/research/gpt-4