AI

Gradium today released two real-time speech translation models: stt-translate again s2s-translate. Both work in five languages ​​and stream results live in the browser. Gradium wants better accuracy for latency tradeoff than gpt-realtime-translate again gemini-3.5-live-translate. It also adds voice output control, including cloning, that gpt-realtime-translate lack. The TL;DR Gradium has introduced two models for real-time speech … Read more

async def demo_memory(): explain( “DEMO 4 — Memory: persistent MEMORY.md across sessions”, “””Long-term memory survives between runs by persisting to MEMORY.md. In session 1 the agent records a user preference; in a brand-new session 2 (fresh engine, fresh transcript) that memory is injected into the system prompt, so the agent already ‘knows’ the user.”””) mem_path … Read more

Most search agents try to handle multiple tasks at once. They create new questions, remember what they’ve already tested, gather evidence, and decide what’s relevant as the search progresses. That can make the whole process messy, expensive, and difficult to manage. Harness-1 takes a simpler approach. Developed by researchers from UIUC, UC Berkeley, and Chroma, … Read more

Nous Research has extended the Skills System within Hermes Agent, its open source self-development agent. The new addition is /learncommand that writes a reusable skill for you. Point to a documentation page, local SDK, past discussion, or sticky notes. The living agent collects the material, then the authors a SKILL.md instead of you. The Hermes … Read more

Large automated language models generate text one token at a time. Each sign waits for the one before it. This serial loop leaves modern GPUs underutilized and keeps processing slow. Costs get worse with longer Chain-of-Thought models. Their long effects make delay a prominent part of the generation. Predictive decoding is a standard fix. A … Read more

Prime Intellect has released prime-rl version 0.6.0. This framework aims to strengthen learning in multi-parameter Mixture-of-Experts (MoE) models. It focuses on agent-heavy tasks, such as long-term software engineering tasks. The research team trained GLM-5 on SWE functions up to 131k sequence length. Step times last less than five minutes. The cluster size was 256 releases. … Read more

xAI has posted a new mod called /goal inside Grok Build, its own code repository agent. The feature is aimed at long-term, automatic operation. You give the agent a lot of initial work, and then back off. Most coding sessions need to be done back and forth and verified. You command, the agent takes action, … Read more

Everyone says Claude can’t make pictures. That’s it in part the truth. Here is the type art it’s self-contained, without plugins and connectors: The sun’s rays turn and the clouds drift away Drawn by Claude in SVG, there is no photo model anywhere near it. Not pixels but code: shapes again links they stay sharp … Read more

The MoonMath AI team released bf16 for AMD’s MI300X GPU. It is written in HIP, not a manual assembly. The code is open source under the MIT license. The MoonMath.ai team reports that it beats AITER v3, AMD’s optimized kernel, in every configuration tested. Bare-metal access came from HotAisle, AMD’s cloud provider. Attention is combined … Read more

import random from collections import Counter, defaultdict from datetime import date, timedelta from prefab_ui.actions import AppendState, OpenLink, PopState, SetState, ShowToast, ToggleState from prefab_ui.app import PrefabApp from prefab_ui.components import ( Alert, AlertDescription, AlertTitle, Badge, Button, Card, CardContent, CardDescription, CardFooter, CardHeader, CardTitle, Code, Column, DataTable, DataTableColumn, Form, Grid, H2, Input, Markdown, Mermaid, Metric, Muted, Progress, Ring, Row, … Read more