Загрузка видео...
Не удалось загрузить видео
David Friedberg says Google basically used to ‘distill’ Microsoft and Yahoo “At Google, in the early days, we would submit millions of search queries to Yahoo and Microsoft search engines to see what the result sets were. We would compare our results against theirs as a way of improving... show more
405,511 просмотров • 1 месяц назад •via X (Twitter)
Комментарии: 34

distillation is a huge problem. I remember at university in the 70s how the professors lectured and students took notes. they were distilling the professors' knowledge! outrage at the stealing! ban note taking! ban attending classes!

It is only illegal if you do it outside of US (and be better)

"But that's impossible! We specifically disallowed that in the terms of service!" - lawyer

Thanks for sharing that. Anyway, I guess the big thing is the geopolitical tension between China and US. Discussion would be different (or non-existent) if everything was among US companies (like Google, Microsoft)

That's just simply benchmarking

ft @friedberg @Jason @theallinpod

Not news to me Sad to see the billionaire geniuses in the Trump admin lacks such common sense

Do you know what “distill” means?

Exactly. Anthropic distills Qwen and other models too.

Absolutely dishonest take. Distillation is a systemic method to reverse engineer the weights of the closed source LLMs, they're trillions times more sophisticated than the search algorithm was 30 years ago. Does Friedberg even have a clue about anything related to AI?

Everybody distills, test each other, benchmarks and learns from each other. That's what competition is. These are all political / regulatory posturing - pre justifications for regulations and demands 😂

Google launched Search in 1998. Microsoft in 2005/2005.

American companies copying each other’s tools is very different than foreign adversaries copying our weapons

What crimes or unethical practices they used, they include attribute them to China. They even have the words for them distillation, overcapacity etc. American firms win by monopolistic practices, stealing technologies and accusing others of stealing. Just losers!

But MSFT was copying and serving results. That's much different than doing competitive analysis.

The most interesting angle is the historical parallel to current AI distillation fights. Companies that now protest others training on their model outputs often used similar output-observation techniques themselves in earlier eras. It helps explain how Google pulled ahead, and it makes modern complaints about distillation look a bit selective.

It's called benchmarking.

Oh damn... You dont say @friedberg ?

Wait are you saying a lot of ai demand is just ai companies distilling eachother

@TaylorLorenz In comparison the Palantir CTO @ssankar was stealing output and replacing it as finished product for prospects when he got caught and was slapped on w/ a RICO IP theft charge. Makes sense to me when comparing the 2

Chik-fil-A distilled Kentucky Fried Chicken?

Microsoft does this heavily to Google today.

distillation is a way of improving existing tools, why do people hate it so much?

This was called comparison shopping in the old days.

Humans learns from each other all the time.

You know, with Google being the undisputed #1 search engine and Bing being a nobody, I would've expected it to be Bing doing all the distilling.

that was then, this is now

As did Microsoft with Google

Would imagine Microsoft and Yahoo! were doing the same

But they were kinda stealing their electricity and bandwidths.

AI is just software. Like a brick it can be used to broke into a house or to build one. The difference between these is the difference between being a criminal or a builder. For anything else blame the AI lab as they won’t take accountability when it’s clearly their fault anyways.

Who hasn’t “distilled” their reading and writing skills the same way?

that competitive analysis approach was pretty standard back then

When china used the same technique americans had flames coming off their arse

