TypeSafe (Organization): news timeline & CVEs

Supersonic Labs Releases Julia 1: A 144.3M-Parameter Open Decision Model That Runs on a CPU

Supersonic Labs released Julia 1, a 144.3M-parameter Apache 2.0 decision model that picks among supplied answers on CPU.

Supersonic Labs, a small Brazilian AI lab, released Julia 1, an open 144.3 million-parameter decision model under Apache 2.0. It takes context, a question, and 2 to 20 candidate answers and returns a choice plus softmax probabilities; it does not generate text. The model keeps JHU CLSP’s mmBERT-small encoder and a decision head; the lab says training and experiments cost about US$104. Pilots beat TypeSafe Jev references on several tasks but scored 64/100 on 72-label Banking77 versus an 87% reference, with a 33.15 ms median per decision on an Apple M4.

MarkTechPostSep 26, 19:50 UTC · 13h agoModel release 2 sources

Ollaya – Ollama for open-source, Jev-style decision models

Ollaya is an open-source local runtime for fast Jev-style decision models, compatible with TypeSafe.

Ollaya is an open-source local runtime for Jev-style decision models, including Convai Innovations' Laya family, that returns calibrated answers in a single forward pass. It reports about 8-10 ms median latency for a five-question HTTP request on an RTX 4090 and serves TypeSafe-compatible /v1/systemone and /v1/models endpoints, so TypeSafe Python SDK 0.7.1 works unchanged. Desktop, CLI, and Docker builds cover macOS, Windows, and Linux under Apache-2.0, with NVIDIA GPU acceleration when CUDA 13 and driver R580 or newer are present.

Hacker News · AISep 25, 18:33 UTC · 1d agoAI tools & infra HN 62↑ · 24 comments

[AINews] The Future of Latent Space

Latent Space outlined business changes and recapped claimed frontier model releases, benchmarks, and a TypeSafe fundraise.

Latent Space said it will merge its Discord and AINews, explore a Beehiiv migration, and reopen sponsorships. Its 23-24 September recap cites Claude Opus 5.5 leading SimpleBench at 88.4% and rising from 24% to 62% on Terminal-Bench-Science at xhigh effort. It also covers a GPT-6 family, Gemini 3.8 Flash scoring 41 on the AA Intelligence Index and 89.2% on ARC-AGI v2, and MIT-licensed Xiaomi MiMo-V2.6-Pro at AA index 46. TypeSafe’s Jev decision model is tied to a reported raise above $1 billion at a valuation above $10 billion, with judging costs far below GPT-6.

Latent SpaceSep 25, 05:37 UTC · 2d agoAI industry 12 sources

What Is RLCD? The Secret Behind Jev

A blog explains Jev's RLCD as calibrated multiway Plackett-Luce preference modeling rather than scalar rewards.

The post explains RLCD as a schema-conditioned Plackett-Luce objective that combines multiway preference modeling with probability calibration. It traces reward modeling from scalar scores to LLaMA-Berry's pairwise preference model, trained on about 7.8 million mathematical-solution pairs, and then to multi-candidate Luce choice. Jev is described as the product form, exposing the reward model through typed outputs rather than hiding it behind a generator. Calibration is framed with negative log loss and the Brier score so reported probabilities match observed accuracy.

Hacker News · AISep 24, 12:21 UTC · 2d agoAI research HN 43↑ · 4 comments

TypeSafe

Timeline4 articles

Appears with

Supersonic Labs Releases Julia 1: A 144.3M-Parameter Open Decision Model That Runs on a CPU

Ollaya – Ollama for open-source, Jev-style decision models

[AINews] The Future of Latent Space

What Is RLCD? The Secret Behind Jev