{"slug": "teach-an-agent-by-doing-the-task-yourself-and-why-the-liquid-net-lost", "title": "Teach an agent by doing the task yourself (and why the liquid net lost)", "summary": "The W3BS Training project shipped a \"Teach an agent\" feature that records a user performing a desktop task, lets them review and redact the recording, and trains a small local model to suggest or, with approval, execute the next step. The release includes the @infernetprotocol/liquid-agent package and a Python runtime supporting CfC and LTC liquid neural networks plus a GRU baseline, with training local by default and opt-in compute workers. On synthetic traces from mock email, Reddit and LinkedIn flows, the GRU outperformed the liquid CfC model on episode success (0.948 vs 0.869), so GRU remains the default; every model failed on unseen screen layouts and none passes the closed-loop safety gate, so the agent refuses to enable itself.", "body_md": "We shipped \"Teach an agent\" in W3BS Training today. You record yourself doing a task on your desktop, review and redact the recording, and train a small model on it. Then the agent can suggest the next step, or do it with your approval.\n\nIt ships in three pieces:\n\n`@infernetprotocol/liquid-agent`, and a Python runtime with three model types: CfC and LTC (the \"liquid\" neural nets) plus a plain GRU baseline. `infernet model train|evaluate|predict|serve` drives it.`--yes`, and the trained weights are checked against the hash in the receipt. Sessions are encrypted end to end.`w3bs training teach`. Training runs locally by default. Sending it to a c0mpute worker is opt-in and shows you a preview first.\nOn a fresh install all five permissions are off and every policy mode is Off. Secrets are masked at capture time, before anything is written. Every external write (sending an email, posting, messaging) needs a person to approve it, in every mode, including Bounded. There is a kill switch and an audit log.\n\nI benchmarked on synthetic traces from mock email, Reddit and LinkedIn flows. The GRU beat the liquid CfC: 0.948 vs 0.869 episode success. So GRU stays the default. Every model failed on a screen layout it had not seen, and nothing passes the closed-loop safety gate yet, so the agent refuses to switch itself on. That is the design working.\n\nThe next step is real recordings of real tasks, not synthetic ones.\n\nStart at [https://w3bs.org/training](https://w3bs.org/training) and open the Teach tab. Everything is off until you turn it on.", "url": "https://wpnews.pro/news/teach-an-agent-by-doing-the-task-yourself-and-why-the-liquid-net-lost", "canonical_source": "https://dev.to/chovy/teach-an-agent-by-doing-the-task-yourself-and-why-the-liquid-net-lost-1p3", "published_at": "2026-10-26 08:40:58+00:00", "updated_at": "2026-10-26 08:43:05.711663+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "machine-learning", "neural-networks", "ai-safety"], "entities": ["W3BS Training", "@infernetprotocol/liquid-agent", "Infernet", "CfC", "LTC", "GRU"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/teach-an-agent-by-doing-the-task-yourself-and-why-the-liquid-net-lost", "markdown": "https://wpnews.pro/news/teach-an-agent-by-doing-the-task-yourself-and-why-the-liquid-net-lost.md", "text": "https://wpnews.pro/news/teach-an-agent-by-doing-the-task-yourself-and-why-the-liquid-net-lost.txt", "jsonld": "https://wpnews.pro/news/teach-an-agent-by-doing-the-task-yourself-and-why-the-liquid-net-lost.jsonld"}}