{"slug": "typesafe-ai-s-meaningful-intelligence", "title": "TypeSafe AI's \"Meaningful Intelligence\"", "summary": "TypeSafe AI introduced a training approach it calls RLCD (reinforcement learning from calibrated decisions), which it says makes models return decisions and probabilities rather than generated text, with outcomes assigned a probability of 0.2 occurring about 20% of the time and outcomes assigned 0.8 occurring about 80% of the time. The company, whose cofounder Diogo Almeida co-invented RLHF used to train InstructGPT and ChatGPT, frames the method as a third post-training path alongside RLHF and RLVR and targets production systems it expects to run about 99% machine-to-machine interactions and 1% human interaction. TypeSafe argues RLHF can reward sycophancy and confident-sounding hallucinations and causes mode dropping, and it published a manifesto describing the approach as Machine Native Intelligence.", "body_md": "**We call this Machine Native Intelligence:**\nAI with software-like properties such as structure, reliability, observability, testability, speed, consistency, and low cost.\n\n## Building prod, not God\n\nTypeSafe is not trying to build a model that does everything. It is designed for production systems where code needs a narrow decision it can inspect and act on. Our expectation is that large-scale AI automation will be closer to 99% machine-to-machine interactions and 1% human interaction. That shifts the design target from responses that feel good to read toward outputs that behave predictably inside software. Read the\n[TypeSafe manifesto](https://typesafe.ai/manifesto).\n\n## Three post-training approaches\n\nPretrained language models have been adapted in two major ways. TypeSafe adds a third. RLHF and RLVR are shown here for context; TypeSafe’s training path is RLCD.\nRLHF was used to train InstructGPT and ChatGPT and was \n\n[co-invented by Diogo Almeida](https://scholar.google.com/citations?user=0T4y07QAAAAJ&hl=en), cofounder of TypeSafe.\n\n## RLCD and calibrated decisions\n\nRLCD optimizes for a different output contract:\n- The model does not generate text.\n- It returns decisions and probabilities.\n- Higher probability should correspond to a greater chance that the answer is correct.\n\n- Outcomes assigned a probability of `0.2` should occur about 20% of the time.\n- Outcomes assigned a probability of `0.8` should occur about 80% of the time.\n- Outcomes assigned a probability of `1.0` should occur 100% of the time.\n\n[Confidence](https://docs.typesafe.ai/confidence)for guidance on deciding when software should act or escalate.\n\n## The problems with RLHF\n\nRLHF teaches a model to say things that people prefer. That objective works well for chatbots, but it can also reward sycophancy and confident-sounding hallucinations. Preference optimization also causes\n**mode dropping**: the model learns to favor a particular style, such as instruction following, while reducing the probability of other possible outputs.\n\n**mode collapse**. In the classic generative-adversarial-network failure mode, a generator learns to produce the same kind of output repeatedly because that output continues to fool the discriminator.\n\n## Mode collapse analogy\n\nMode collapse analogy", "url": "https://wpnews.pro/news/typesafe-ai-s-meaningful-intelligence", "canonical_source": "https://docs.typesafe.ai/introduction/machine-learning-primer", "published_at": "2026-09-18 10:40:19+00:00", "updated_at": "2026-09-18 10:55:40.021744+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "large-language-models", "ai-research", "ai-agents"], "entities": ["TypeSafe AI", "RLCD", "RLHF", "RLVR", "Diogo Almeida", "InstructGPT", "ChatGPT"], "alternates": {"html": "https://wpnews.pro/news/typesafe-ai-s-meaningful-intelligence", "markdown": "https://wpnews.pro/news/typesafe-ai-s-meaningful-intelligence.md", "text": "https://wpnews.pro/news/typesafe-ai-s-meaningful-intelligence.txt", "jsonld": "https://wpnews.pro/news/typesafe-ai-s-meaningful-intelligence.jsonld"}}