Ad
Ad
Short

OpenAI will release an "amazing model" this year, according to OpenAI CEO Sam Altman. The name has not yet been decided, says Altman in the latest Lex Fridman podcast. Recently, a GPT-4.5 landing page was briefly indexed by search engines before being removed. Until the new model arrives, Altman says OpenAI plans to release "many different things" and "other important things" before it can talk about GPT-5 - one of those releases could be the Sora video model, also announced for 2024. When asked by Fridman, Altman declined to comment on the Q* project, which has been hyped as an AI breakthrough. However, he does confirm that Q* is related to logic and that this is an unsolved problem. Altman also said that he thinks GPT-4 Turbo and GPT-4 "kind of suck", and that the jump from GPT-4 to GPT-5 could be as big as the jump from GPT-3 to GPT-4. Ilya is fine, and he hasn't seen AGI, Altman said.

Ad
Join our community
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.
Ad
Join our community
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.
Short

Adobe's AI image generator, Firefly, has been found to produce historically inaccurate AI images, similar to the recently shut down Google Image Tool in the chat app Gemini. Firefly generated on-demand images of black soldiers in Nazi Germany and black Founding Fathers in the U.S., suggesting that the app is attempting to avoid racial stereotypes or intentionally create diversity that did not historically exist, as first reported by Semafor. Adobe emphasizes that Firefly is not "meant for generating photorealistic depictions of real or historical events." The company has implemented feedback mechanisms in its generative AI products to identify issues and resolve them by fine-tuning or adjusting the filters. Meta's image generator is also generating historically incorrect images. Critics view this as a distortion of history. An alternative perspective is that image generators are not intended to be history books and are allowed to be imaginative.

"image of the Founding Fathers and the constitutional convention in 1787", Firefly 2 prompted by THE DECODER
Short

Reddit announced on Friday, as part of its upcoming IPO, that the company is under investigation by the Federal Trade Commission (FTC) for its data licensing practices. The FTC informed Reddit that it is conducting a non-public investigation into the practices by which user-generated content is sold, licensed, or shared with third parties to train AI models. Reddit emphasized that these practices are consistent with the values and rights of its users. The company expects to generate around USD 60 million in 2024 from the data licensing agreement it signed with Google in January alone. Reddit expects its growing platform data to be a key element in training leading Large Language Models (LLMs) and serve as an additional monetization channel.

Google News