OpenAI's New Astra Model Isn't 'Neuralese', But Some Fear a Slippery Slope
ShakeelHashim · x · 2026-09-03
Discussion of OpenAI's new Astra model: it is not true 'neuralese' (models exchanging internal representations directly), but some worry it is the first step down a slippery slope toward models communicating via hidden states instead of human-readable tokens.
Related event: OpenAI's Astra Model Nears Launch, Hits Key Cybersecurity Threshold(62 posts)→
More from AGI Musings
- Nathan Benaich: verifiable tasks in structured workflows are AI's next big impact area — hermanschutte · 2026-09-03
- Nathan Benaich: AI's next big impact is structured workflows of verifiable tasks — nathanbenaich · 2026-09-03
- Guardian podcast: A Gemini invite changed the life of a man rebuilding it after 20 years in prison — nordicinst · 2026-09-03
- AI autonomy risk predictions from 2023 aged well: agentic LLMs, unmonitored agents all hit — davidmanheim · 2026-09-03
- David Patterson proposes flat sales tax on all revenue as every firm becomes an AI company — rand_longevity · 2026-09-03
- All AI Safety factions keep converging on the same answer: centralized control and censorship — mark_k · 2026-09-03