4 ms·

Seen enough cycles of laying off about 10% [1] of your workforce to try and halt a share price slide and make investors happier to realise it's in the Howto CEO

Seen enough cycles of laying off about 10% [1] of your workforce to try and halt a share price slide and make investors happier to realise it's in the Howto CEO & Keep Your Job manual. QED, this announcement is almost entirely about share prices sliding in the tech sector since about the beginning of 2022. ChatGPT's impact on Google is unknown, but it is a fair bet that it will be larger [2]. [1] It's always around 10% because all large companies carry functions and individuals who can be removed from the enterprise, and things will essentially carry on as before once the shock subsides. [2] I don't think many would doubt Google could have rolled out its own ChatGPT'alike years ago. The fact that they didn't suggests that internally a combination of the undoubted reputational risk, cost and the risk to their existing revenue streams was enough to prevent it. Google's full of very bright people, and they've failed to find a palatable solution in years. So even if it is a solvable problem, it seems unlikely that the recent nasty shock from ChatGPT is going to have changed that.

Neil44 4y ago
Despite the hype ChatGPT is not ready to replace search engines yet - no referencing and it literally makes up gibberish if it's not sure. At this point I don't see any advantage to Google rolling that out. GPT4, GPT5... different story maybe!
- visarga 4y ago
  The hallucination part is somewhat solvable, but the inference cost is huge. Especially in the chat format where you want long history, you have to pay for all the previous tokens in each round of dialogue. The cost is about $0.01 per page, so it can become $0.10 per reply after 10 pages of history accumulated. Even short lines of text have the same overhead.
- lonelyasacloud 4y ago
  Fundamentally a Google search typically returns a research project that involves the user attempting to build their own mental model of the "truth". ChatGPT instead returns an answer that needs research to verify facts. This second approach is already quicker for many use cases and can get much better. So yes, the user may still need search engines to verify. But in the future, it seems unlikely that search will be the dominant, primary means of people accessing information. The problem for Google with ChatGPT-a-like results is that they strip out the presentation of Google's ads and one of its most significant revenue streams. The chief advantage of Google acting quickly is it increases its chances of retaining an ad business.

Neil44 4y ago

Despite the hype ChatGPT is not ready to replace search engines yet - no referencing and it literally makes up gibberish if it's not sure. At this point I don't see any advantage to Google rolling that out. GPT4, GPT5... different story maybe!

visarga 4y ago
The hallucination part is somewhat solvable, but the inference cost is huge. Especially in the chat format where you want long history, you have to pay for all the previous tokens in each round of dialogue. The cost is about $0.01 per page, so it can become $0.10 per reply after 10 pages of history accumulated. Even short lines of text have the same overhead.
lonelyasacloud 4y ago
Fundamentally a Google search typically returns a research project that involves the user attempting to build their own mental model of the "truth". ChatGPT instead returns an answer that needs research to verify facts. This second approach is already quicker for many use cases and can get much better. So yes, the user may still need search engines to verify. But in the future, it seems unlikely that search will be the dominant, primary means of people accessing information. The problem for Google with ChatGPT-a-like results is that they strip out the presentation of Google's ads and one of its most significant revenue streams. The chief advantage of Google acting quickly is it increases its chances of retaining an ad business.