llm - Tech News from Simmons Systems (Page 8)

Tech News from Simmons Systems

llm

A collection of 153 posts

OpenAI API Now Guarantees Valid JSON with Structured Outputs

OpenAI introduces a new Structured Outputs feature for its API that guarantees valid JSON responses matching specific schemas. This update eliminates the previous unreliability where models occasionally returned malformed data.

Google's AI Search Feature Struggles With Inaccurate And Bizarre Answers

Google's new AI Overviews search feature faces heavy criticism for generating bizarre and potentially dangerous recommendations, such as eating rocks. The company is now scrambling to implement technical fixes to limit the inclusion of misleading online content.

OpenAI Upgrades ChatGPT Premium With Smarter GPT-4 Turbo Model

OpenAI releases an improved GPT-4 Turbo model for its paid ChatGPT tiers, offering better performance in writing, math, and coding. The update extends the model's knowledge base to December 2023 and shows significant accuracy gains on complex benchmarks.

Spotify Introduces AI Playlists Built from Written Prompts

Spotify releases a beta feature that lets users create custom playlists using written prompts, leveraging large language models and personalization technology. The tool is currently available on mobile devices in the U.K. and Australia.

Google Launches Gemma, Open-Source AI Models Based on Gemini Tech

Google introduces Gemma, a lightweight open-source AI model family built on Gemini research. The models allow developers to run capable AI directly on laptops without needing expensive cloud access.

Google Launches Gemini 1.5 Pro With Unprecedented One Million Token Context Window

Google introduces the Gemini 1.5 Pro AI model featuring a massive one million token context window capable of processing entire books or hours of video. The experimental model currently faces latency issues but offers a major leap in long-context understanding.

Google Unveils Gemini, a Multi-Modal AI Model Surpassing Human Experts

Alphabet releases Gemini, a new multi-modal AI model developed by Google DeepMind that outperforms human experts on key language understanding benchmarks. The model launches in three distinct sizes to power various Google products and developer tools.

Mistral AI Unveils Mixtral 8x7B to Advance Open AI Models

Mistral AI releases Mixtral 8x7B, a new open-weights model utilizing a sparse mixture-of-experts architecture to push the boundaries of artificial intelligence.

Google Upgrades Bard Chatbot With New Gemini AI Model

Google integrates its new Gemini AI model into Bard, promising major improvements in reasoning and planning. The update initially brings a tailored Gemini Pro version to the chatbot, with a more powerful Ultra model planned for next year.

Google Pours $2 Billion Into Anthropic Amid Intense AI Proxy War

Google reportedly invests $2 billion into Anthropic, highlighting a fierce proxy war among tech giants who cannot build their own large language models fast enough.

Copyright Lawsuits Mount as EU AI Act Negotiations Face Delays

A new wave of federal copyright lawsuits targets major AI developers over the use of copyrighted books and song lyrics in training data. Meanwhile, disagreements among EU lawmakers threaten to push final negotiations for the AI Act into 2024.

Apple to Spend Billions on AI Servers to Train Language Models

Analyst Ming-Chi Kuo predicts Apple will spend $4.75 billion on AI servers in 2024 to build its own large language models. This massive investment highlights a race to catch up with competitors like Microsoft and Meta.

Mistral AI Unveils Efficient 7.3B Open-Source Language Model

Generative AI startup Mistral releases a highly efficient 7.3 billion parameter language model under an Apache 2.0 license to challenge larger corporate AI systems.

Mistral AI Launches Open-Source 7B Parameter Language Model Ahead of Schedule

Paris-based startup Mistral AI releases Mistral 7B, a highly capable open-source language model available for commercial use under an Apache 2.0 license. The new model outperforms competing open models up to 13 billion parameters on standard benchmarks.