Model
gpt-4
GPT-4 was OpenAI's flagship model from March 2023, and the first widely used system that could read images as well as text while scoring near the top of the human range on many professional exams. Its technical report broke with earlier practice by disclosing nothing about how the model was built — its size, its data or its training — which OpenAI defended on competition and safety grounds and critics read as the end of the field's openness. For roughly two years it was the model others were measured against, until newer multimodal and reasoning systems succeeded it.
Appears alongside
Featured in threads
Tracks
- Models & capabilities 7
- Courts & copyright 5
- Money & business 4
- Safety & alignment 3
- Ideas & essays 2
- Benchmarks & progress 2
- Culture & impact 2
- Security & misuse 1
- Open weights & ecosystem 1
Munich court rules OpenAI infringed German song lyric copyrights
The court found that lyrics retained in a model's parameters count as an infringing reproduction, rejecting OpenAI's text-and-data-mining defence; OpenAI may appeal.
Courts & copyright
OpenAI publishes early sparse-autoencoder work extracting concepts from GPT-4
The 16-million-feature autoencoder cost roughly as much accuracy as training GPT-4 with ten times less compute, illustrating interpretability's overhead at scale.
Safety & alignment
Judge dismisses most of Silverman's book-author lawsuit against OpenAI
The judge found the authors had not shown ChatGPT's output resembled their books closely enough, but let the core training-data infringement claim proceed.
Courts & copyright
OpenAI opens the GPT Store
A marketplace for custom assistants, launched two months late after the board crisis, alongside a new $25-a-month ChatGPT Team subscription tier.
Models & capabilities · Money & business
The New York Times sues OpenAI and Microsoft
The first major news organisation to sue over AI training data, alleging models could reproduce its articles near-verbatim.
Courts & copyright
OpenAI publishes 'Weak-to-Strong Generalization' superalignment paper
Fine-tuning GPT-4 on labels from a GPT-2-sized supervisor recovered close to GPT-3.5-level performance on language tasks, but the technique still struggled on chess puzzles and reward modelling.
Ideas & essays · Safety & alignment
Researchers show GPT-4 safety guardrails easily bypassed via low-resource languages
Researchers traced the gap to RLHF: human annotators who flag harmful content overwhelmingly work in high-resource languages, leaving weaker refusal training elsewhere.
Security & misuse
ChatGPT gains voice and vision
A new text-to-speech model, built with professional voice actors, let users talk to ChatGPT and show it photos, ahead of the fully real-time GPT-4o release.
Models & capabilities
The Authors Guild sues OpenAI
Seventeen novelists including Grisham and Martin filed a class action in the Southern District of New York alleging systematic copying of pirated books to train GPT.
Courts & copyright
OpenAI launches ChatGPT Enterprise
The tier offered unlimited GPT-4 access at double speed, a 32,000-token context window and a pledge not to train on business data.
Models & capabilities · Money & business
Authors begin suing over training data
Comedian Sarah Silverman and novelists Richard Kadrey and Christopher Golden filed separate suits in California, seeking class-action status against both companies.
Courts & copyright
Tree of Thoughts adds search to model reasoning
Letting a model branch, evaluate and backtrack over intermediate steps raised Game-of-24 success from 4% to 74% against chain-of-thought.
Ideas & essays · Benchmarks & progress
AutoGPT starts the agent craze
The open-source project let GPT-4 set and pursue its own subgoals with no human in the loop, but often stalled in expensive, unproductive repetition.
Open weights & ecosystem
ChatGPT gets plugins and browsing
Initial plugins let ChatGPT book flights, order groceries and query Wolfram Alpha through partners including Expedia, Instacart, Klarna and Zapier.
Models & capabilities
Microsoft announces 365 Copilot
Built on GPT-4 and Microsoft's own data graph, it launched to a small group of testers with no price or general release date set.
Money & business · Culture & impact
OpenAI releases GPT-4
A multimodal model that passed professional exams near the top of the human range — and whose technical report disclosed no architecture, data or compute.
Models & capabilities · Benchmarks & progress
Bing's chatbot tells a journalist to leave his wife
Days after the exchange, Microsoft capped Bing chat sessions to five turns, saying long conversations could 'confuse' the model into drifting from grounded answers.
Culture & impact · Safety & alignment
Microsoft puts GPT-4 inside Bing
Microsoft called it only a 'next-generation OpenAI large language model'; the company confirmed five weeks later, on GPT-4's public release, that Bing had been running on GPT-4 all along.
Models & capabilities · Money & business
Harvey signs exclusive launch partnership with Allen & Overy
More than 3,500 lawyers at 43 offices got access to the GPT-4-based tool after a beta that logged roughly 40,000 queries since November 2022.
Models & capabilities