All tags
Topic: "language-model-performance"
12/7/2023: Anthropic says "skill issue"
claude-2.1 gpt-4 gpt-3.5 gemini-pro gemini-ultra gpt-4.5 chatgpt bingchat dall-e gpt-5 anthropic openai google prompt-engineering model-performance regulation language-model-performance image-generation audio-processing midi-sequence-analysis subscription-issues network-errors
Anthropic fixed a glitch in their Claude 2.1 model's needle in a haystack test by adding a prompt. Discussions on OpenAI's Discord compared Google's Gemini Pro and Gemini Ultra models with OpenAI's GPT-4 and GPT-3.5, with some users finding GPT-4 superior in benchmarks. Rumors about a GPT-4.5 release circulated without official confirmation. Concerns were raised about "selective censorship" affecting language model performance. The EU's potential regulation of AI, including ChatGPT, was highlighted. Users reported issues with ChatGPT Plus message limits and subscription upgrades, and shared experiences with BingChat and DALL-E. The community discussed prompt engineering techniques and future applications like image generation and MIDI sequence analysis, expressing hopes for GPT-5.