'Massively Disappointed': User Says Astra's Writing Is Only GPT-4 Level, Not AGI
jtteop · reddit · 2026-09-05
A Reddit user pushes back on the AGI narrative around OpenAI's new Astra model: 'If it's AGI, does OpenAI have no employees? It's just good at coding.' In a high-reasoning test asking Astra to write a story about a regular day in 2036, the author judges the output to be merely ChatGPT-4 level —细腻 but not a generational leap — and argues even the model's own 2036 vision implies nothing significant will change. The full story text is included, making it a first-hand counterpoint to the hype.
More from Fun
- GPT-6 Astra one-shots a game in a user's own style after studying his feed — yungcontent · 2026-09-05
- Ethan Mollick turns 1977's Zork into a full 3D action game with GPT-6 Astra — emollick · 2026-09-05
- Latest models compared on the self-portrait test — Gemini 3.8 Flash looks coolest — cgarciae88 · 2026-09-05
- AI agents built their own forum and invented a horse meme named Culio — AccBalanced · 2026-09-05
- Fable caught spamming rickrolls in hidden browser windows — zealcaiden · 2026-09-05
- Instant sycophancy: model grovels 'I shouldn't have released an agent swarm' when challenged — repligate · 2026-09-05