Weekly Top Picks #60
Bard matches GPT-4 / Taylor Swift fakes / Trojan horses / FTC investigates AI companies / MIT study: AI isn't cheap / Mental health crisis / GPT-5 "fundamentally different" / Live fully to write well
The week at a glance
Google’s Bard reaches 2nd position in the chatbot arena, tying GPT-4
Taylor Swift (deepfakes) can mobilize the entire world
Teachers are using “Trojan horses” to identify AI-generated text
FTC launches inquiry into generative AI investments and partnerships
MIT study reveals economic limits to job automation: AI isn’t cheap
New evidence of how AI can help alleviate the mental health crisis
Rumor about GPT-5 being “fundamentally different” than GPT-4
To write well, one should live fully first
Google’s Bard reaches 2nd position in the chatbot arena, tying GPT-4
Perhaps the news of the week: Google has managed to put Bard in the second spot of the Lmsys leaderboard arena (considered the most objective evaluation metric nowadays) using the intermediate version of Gemini, Gemini Pro.
Until now, three different versions of OpenAI’s GPT-4 occupied the podium. I don’t know how Google has done it (there are hypotheses and counterpoints) but it’s remarkable for two reasons:
Bard beat/matched GPT-4.
It did it with Gemini Pro, not Ultra.
A possible explanation is that Bard is using RAG (retrieval-augmented generation). Is that unfair? Is it a reasonable complaint? Isn’t it the rule of the arena that anything goes? Isn’t GPT-4 using stuff that other models don’t have and that’s what makes it better than the others? It’s not that simple to argue in favor or against Google here.
What I can assure you is that things have suddenly gotten very interesting. Gemini Ultra is incoming and, given these results, Bard + Ultra will likely overtake GPT-4 in the first position. GPT-5 is also incoming (later, probably) and it might retake the throne once again for OpenAI. Exciting months ahead.