# forck.live — 16–22 June 2025 > AI lab announcements, news reports and podcasts. forck.live follows what companies announce about their own products. Press may corroborate an official story; it never leads and never appears as an item of its own. A missing date stays missing. A title is the publisher's words — where a company does not publish in English, an English rendering is listed separately as `Title in English` and named as this site's translation, never as theirs. This file is one issue of the weekly archive: every confirmed first-party announcement of ISO week 2025-W25, 16–22 June 2025, newest first. The window never moves, so this URL always names the same seven days — but a post published inside them and detected later is added when it arrives, so a closed week can still gain entries. Every entry is one first-party announcement, and carries the same eight fields in this order before any others: Company · Announced · Category · Coverage · Announcement · Group · Source · Record `Source` is the company's own page the announcement was traced to and `Record` is this site's page for it. Both are safe to cite. `Coverage` is a number of outlets, or `not counted` — which means this record reached no count of its own, never that nobody wrote about it. `Category` is this tracker's own analysis, and is authoritative wherever it is stated. `not stated` means either that nothing has analysed the post yet or that the analysis placed it outside this taxonomy; neither is a verdict about the announcement. `Announcement` is whether the post is news of something the company did, rather than a how-to, a customer story, a changelog listing or a staffing note. `Group` is which of this record's groups the entry falls in, carried on the entry itself so it holds where the groups are headings and where they are not. One of: `models`, `covered`, `announcements`, `routine`. Anything after `Record` is held detail that only some announcements carry. - This week's page: https://forck.live/week/2025-W25 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### Sakana AI — Reinforcement Learning Teachers of Test Time Scaling - Company: Sakana AI (sakana.ai) - Announced: 2025-06-22T15:00:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://sakana.ai/rlt - Record: https://forck.live/items/4713-reinforcement-learning-teachers-of-test-time-scaling - Subject: Sakana models Sakana AI introduces Reinforcement-Learned Teachers (RLTs), a new method for training teacher models to generate explanations for student models by learning to teach rather than solve problems. The teacher is rewarded based on the student's understanding. Small 7B parameter RLTs outperform much larger models like DeepSeek R1 (671B) in teaching reasoning skills. The code, paper, and open models are released. ### Kyutai — Kyutai Speech-To-Text released as open-source - Company: Kyutai (kyutai.org) - Announced: 2025-06-19 - Category: open-weight-release - Coverage: not counted - Announcement: yes - Group: models - Source: https://kyutai.org/blog/2025-06-19-stt-open-source - Record: https://forck.live/items/18354-kyutai-speech-to-text-released-as-open-source - Subject: Moshi / Pocket TTS Kyutai released its streaming speech-to-text model, Kyutai STT, as open-source. The model powers the company's Unmute product. ### Hugging Face — (LoRA) Fine-Tuning FLUX.1-dev on Consumer Hardware - Company: Hugging Face (huggingface.co) - Announced: 2025-06-19 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/flux-qlora - Record: https://forck.live/items/1674-lora-fine-tuning-flux-1-dev-on-consumer-hardware - Subject: Platform This blog post guides users through fine-tuning the FLUX.1-dev model using QLoRA with the diffusers library, achieving peak memory usage under ~10 GB VRAM on a single GPU like the NVIDIA RTX 4090. It covers techniques such as 4-bit quantization, 8-bit AdamW optimizer, gradient checkpointing, and latent caching to enable efficient fine-tuning on consumer hardware. ### Replit — Introducing Effort-Based Pricing for Replit Agent - Company: Replit (replit.com) - Announced: 2025-06-18T17:02:11+00:00 - Category: pricing-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/effort-based-pricing - Record: https://forck.live/items/17963-introducing-effort-based-pricing-for-replit-agent - Subject: Replit Agent Replit replaced its $0.25-per-checkpoint pricing for Replit Agent with an Effort-Based pricing model that charges per request based on the time and computation used. Simple tasks now typically cost less than $0.25, while complex tasks are bundled into a single checkpoint that may cost more. The new pricing is effective immediately for new users and rolls out to existing Core and Teams subscribers starting July 1st, alongside new per-request controls for a high-power model and extended thinking. ### OpenAI — Toward understanding and preventing misalignment generalization - Company: OpenAI (openai.com) - Announced: 2025-06-18T10:00:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/emergent-misalignment - Record: https://forck.live/items/538-toward-understanding-and-preventing-misalignment-generalization - Subject: GPT / ChatGPT / API The study examines how training on incorrect responses leads to broader misalignment in language models and identifies an internal feature that drives this behavior, which can be reversed with minimal fine-tuning. ### OpenAI — Preparing for future AI risks in biology - Company: OpenAI (openai.com) - Announced: 2025-06-18T10:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/preparing-for-future-ai-capabilities-in-biology - Record: https://forck.live/items/537-preparing-for-future-ai-risks-in-biology - Subject: GPT / ChatGPT / API OpenAI is proactively assessing capabilities and implementing safeguards to prevent misuse of advanced AI in biology and medicine due to biosecurity risks. ### Cua — Your Windows PC is Already the Perfect Development Environment for Computer-Use Agents - Company: Cua (cua.ai) - Announced: 2025-06-18 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://cua.ai/blog/windows-sandbox - Record: https://forck.live/items/16839-your-windows-pc-is-already-the-perfect-development-environment-for-computer - Subject: Cua Driver / Lume / Cua Bench / Cua Fleets Cua added support for Windows Sandbox as a local development environment for computer-use agents, enabling users to test and develop agent workflows on Windows 10/11 without cloud infrastructure. The company plans to release Windows cloud instances later in June 2025 for production workloads. ### Harvey — Low Latency Redline Detection via Vision Models - Company: Harvey (harvey.ai) - Announced: 2025-06-17T16:00:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.harvey.ai/blog/low-latency-redline-detection - Record: https://forck.live/items/17707-low-latency-redline-detection-via-vision-models - Subject: Harvey Harvey developed a custom vision model to detect redlined documents in PDFs, replacing slower text-based heuristics. The model uses a lightweight CNN with 100K parameters, achieving sub-500 ms CPU inference and a validation F1 of 0.992. This allows Harvey to selectively apply intensive document processing only when redlines are present, reducing latency and improving accuracy. ### Isomorphic Labs — Isomorphic Labs appoints Dr. Ben Wolf as Chief Medical Officer and establishes US Presence - Company: Isomorphic Labs (isomorphiclabs.com) - Announced: 2025-06-17 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.isomorphiclabs.com/articles/isomorphic-labs-appoints-dr-ben-wolf-as-chief-medical-officer-and-establishes-us-presence - Record: https://forck.live/items/19136-isomorphic-labs-appoints-dr-ben-wolf-as-chief-medical-officer-and-establishes - Subject: Drug Design Engine / AlphaFold Isomorphic Labs appointed Dr. Ben Wolf as Chief Medical Officer and established a US presence with a new Cambridge location, where he will build a biomedical team to advance AI-driven drug discovery. ### Sakana AI — Towards Automating Long-Horizon Algorithm Engineering for Hard Optimization Problems - Company: Sakana AI (sakana.ai) - Announced: 2025-06-16T15:00:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://sakana.ai/ale-bench - Record: https://forck.live/items/4714-towards-automating-long-horizon-algorithm-engineering-for-hard-optimization - Subject: Sakana models Sakana AI introduces ALE-Bench, a coding benchmark for hard optimization problems, and ALE-Agent, a coding agent that achieved 21st place among over 1,000 human participants in a live AtCoder Heuristic Competition. ### Vercel — Tray.ai cut build times from a day to minutes with Vercel - Company: Vercel (vercel.com) - Announced: 2025-06-16T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/tray-ai-cut-build-times-from-a-day-to-minutes-with-vercel - Record: https://forck.live/items/20862-tray-ai-cut-build-times-from-a-day-to-minutes-with-vercel - Subject: v0 / AI SDK / AI Gateway Tray.ai, a composable AI integration and automation platform, partnered with Roboto Studio to migrate their marketing site to Vercel. The migration reduced build times from a full day to two minutes, consolidated multiple repos into a single Turborepo monorepo, and improved developer experience with preview deployments and real-time analytics. ### Hugging Face — Groq on Hugging Face Inference Providers 🔥 - Company: Hugging Face (huggingface.co) - Announced: 2025-06-16 - Category: api-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/inference-providers-groq - Record: https://forck.live/items/1675-groq-on-hugging-face-inference-providers - Subject: Platform Groq is now a supported Inference Provider on the Hugging Face Hub, offering fast LPU-powered inference for open-source models like Llama 4 and QWQ-32B, with integration into the Hub's UI and client SDKs. ### OpenAI — Introducing OpenAI for Government - Company: OpenAI (openai.com) - Announced: 2025-06-16 - Category: availability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/global-affairs/introducing-openai-for-government - Record: https://forck.live/items/539-introducing-openai-for-government - Subject: GPT / ChatGPT / API OpenAI launches OpenAI for Government, an initiative to bring advanced AI tools to U.S. public servants.