I don’t think this is evidence of cached responses; the logprobs may just be that high. GPT has had issues with joke reuse at least since RLHF (and likely before with FeedMe) — e.g. ChatGPT (GPT-3.5) was found to reuse the same 25 jokes for 90% responses: https://
arxiv.org/abs/2306.04563
AI
-
Goodside: high logprobs not cached responses cause joke reuse
By
–
-
AMD Poised for Major AI Growth in 2024
By
–
AMD could have quite an AI year in 2024 Advanced Micro Devices indicated its data center sales could surge in 2024 by almost ten-fold. This would take the company’s business to a new level as it would bring direct, meaningful competition to Nvidia for artificial intelligence.
-
AI Pest Management Helps Indian Farmers Increase Profitability
By
–
Nakul Jain, our Director of Solutions, discusses our AI impact work which helps Indian farmers make more profitable decisions and the valuable support we have received from @Googleorg to scale our pest management technology with @AgriGoI
. @business_today -
GEN:48 Grand Prix Winners Announced: Dan Hammill and Jeff Wood
By
–
Tens of thousands of votes, 65 finalists, one Grand Prix winner.
— Runway (@runwayml) 1 novembre 2023
Congratulations to Dan Hammill and Jeff Wood for winning the 1st edition of GEN:48 with their film, "2026."
If you haven't already, be sure to watch the film (alongside all other finalists) at… pic.twitter.com/iQZUg1ivDxTens of thousands of votes, 65 finalists, one Grand Prix winner. Congratulations to Dan Hammill and Jeff Wood for winning the 1st edition of GEN:48 with their film, "2026." If you haven't already, be sure to watch the film (alongside all other finalists) at
-
IBM Toronto Data Centre 1963: Computing History Milestone
By
–
Fall’n down memory lane Pictured below: IBM Toronto Data Centre, 1963
-
Podcast explores AI thriller and real-world scientist LARP phenomenon
By
–
These posts brought to you by a Marketplace Tech podcast episode covering someone LARPing as a scientist in an AI thriller (unfortunately, well-funded in the real-world), which ended with a pointer to this earlier interview:
-
Ahistoricism in AI Coverage: Focus on Actual Harms
By
–
Second, this blog post regarding ahistoricism in some of the coverage and the importance of keeping the focus on actual harms and the activists and scholars who are working on addressing them: https://
medium.com/@emilymenonben
der/talking-about-a-schism-is-ahistorical-3c454a77220f
… >> -
AI Safety Concerns and Response to AI Pause Letter
By
–
In light of all the extra AI doomerism in the news (likely due in part to the "AI Safety Summit" in the UK this week), I'd like to re-up a couple of resources. First, this response to the "AI Pause" letter: https://
dair-institute.org/blog/letter-st
atement-March2023/
… >> -
Mozilla Foundation Letter Advocates for Open AI Platforms
By
–
Openness, transparency, and broad access makes software platforms safer and more secure.
This open letter from the Mozilla Foundation, which I signed, makes the case for open AI platforms and systems. -
Open Source Platforms Enhance AI Safety and Security
By
–
Open source platforms *increase* safety and security.
This is as true for AI as it is for operating systems and Internet infrastructure software.