The majority of what you see on SERPs has been AI generated for a decade, and that's Google's main money maker.
This idea that Google sucks at AI and is clearly a million miles behind is intriguing.
Google is currently the company everyone loves to hate like MSFT was 5-10 years ago.
THAT is a bad sign regardless if anything the people are saying has any truth to it.
MSFT has been the king of enterprise software since the 80s. But around 2000, everyone and their mother was talking about how MSFT can't produce any software beside complete garbage.
None of that was true. But, MSFT was in for some relatively bad times for a decade or so.
I suspect the same thing is going to happen to Google just because sentiment is powerful.
Google was good at attracting and retaining the best and brightest. Now they aren’t.
“Based in part on his work with Hinton, Sutskever was hired by Google. There, he implemented a neural-network-driven approach to language translation that produced fewer errors than competing efforts. His work provided the basis for a major upgrade to Google Translate. “Researchers didn’t believe that neural networks could do translation, so it was a big surprise when they could,” he says.”
https://magazine.utoronto.ca/people/alumni-donors/heard-it-t...
Hinton left last week as well.
And they had their lunch completely eaten by deepL
Ummm… In 2000s MSFT was producing complete garbage. Internet Explorer, Windows XP then ME, Windows Phone, ASP, Bing.
It's also the single clearest example of a product that's been left behind by the "AI revolution" - while Google may still have the numbers (like Yahoo Search did back in the day), DeepL is what serious people use.
> The majority of what you see on SERPs has been AI generated for a decade, and that's Google's main money maker.
It's also got subjectively worse and worse over that same time period. Demand for a better search engine is higher than it's been for decades.
I tested GPT 3.5, GPT 4.0, Google Translate, and DeepL on some "difficult" sentences, and GPT 4 was the only one that could translate them correctly. Not only that, it could do style shifts in non-English languages, such as "make this more formal", "less formal", "idiomatic", etc...
I suspect that GPT 4 is so good at translation is because it understands what it is translating at a near-human level.
I like this snippet:
"John Bear from Bear Stearns was bearish about bear pelt futures on the Bear City bear markets."
Which GPT 4 can round-trip from English -> Hungarian -> Chinese -> English again and outputs this:
"John Bell comes from Bellsterns and has a very strong bearish sentiment on the future value of bear skins in the Bear City bear market."
Interesting to note that it used a sounds-alike-to-Chinese-speakers "Bell" for "Bear" in proper names. Other than that, this is pretty good, which is very impressive. DeepL and Google Translate shreds the first step and have no hope on a round-trip this long across such unrelated languages. For example, DeepL outputs some gibberish about bear claws (the pastry) and bear cubs!?
https://www.techradar.com/features/google-glass-2-everything...
Apple is apparently announcing their AR glasses in June
https://www.bloomberg.com/news/newsletters/2023-04-02/when-i...
Netflix thought they had a similar moat of user-habits. But it turns out that the more important moat in the streaming market is a moat of content.
I really doubt that. If Google had anything that could seriously compete with GPT-4 they would be shouting about it from the mountains. Not publishing a comparison where they only include GPT-4 in the two benchmarks that they win.
If that is really the best that Google can do after working for several months on one-upping GPT-4, that's just sad, and doesn't bode well for them.
Deepmind has probably the best research in the AI area fundamentally, and the capabilities they have shown academically are much better than any other group. However, they have been seemingly completely disinterested in making any product usable for the public in the area - which takes a good deal of engineering and glue. So it will take a bit of time for them to make that happen. For example, OpenAI was not the first to use RLHF to train a language model (Sparrow from Deepmind was the first to conceive of this idea). Additionally, Deepmind has far better people in the general GNN and algorithmic reasoning area, which is going to shape up to be far more important in the coming years. That's of course not to say OpenAI doesn't have excellent research or people (Sutskyver is integral to an incredible number of monumental achievements in NNs), but the company as a whole focuses on product far more than fundamental academic research, and that has a potential to creep up on them.
Obviously Bard from the scientific side seems like it was probably hastily thrown together in an afternoon by one guy, and then sent to the marketing/we dev/etc group to put it public facing.
Good thing OpenAI has a distribution partner in Microsoft Azure, then.
Dude, you were seated in the pot of gold and let your organizational disfuncionality block you from really using that.
And now you come trying to get the crown of AI? Catch up and them we can talk.