# forck.live — 20–26 January 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-W04, 20–26 January 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-W04 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### Alibaba — Qwen2.5-1M: Deploy Your Own Qwen with Context Length up to 1M Tokens - Company: Alibaba (alibaba.com) - Announced: 2025-01-26T16:00:03+00:00 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://qwenlm.github.io/blog/qwen2.5-1m/ - Record: https://forck.live/items/2297-qwen2-5-1m-deploy-your-own-qwen-with-context-length-up-to-1m-tokens - Subject: Qwen Alibaba's Qwen team released two open-source models, Qwen2.5-7B-Instruct-1M and Qwen2.5-14B-Instruct-1M, which support a context length of up to 1 million tokens. ### Alibaba — Qwen2.5 VL! Qwen2.5 VL! Qwen2.5 VL! - Company: Alibaba (alibaba.com) - Announced: 2025-01-26T11:08:30+00:00 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://qwenlm.github.io/blog/qwen2.5-vl/ - Record: https://forck.live/items/2298-qwen2-5-vl-qwen2-5-vl-qwen2-5-vl - Subject: Qwen Alibaba releases Qwen2.5-VL, a new vision-language model, and opens base and instruct models in three sizes (3B, 7B, 72B) on Hugging Face and ModelScope. ### Lovable — How to integrate Replicate with Lovable? - Company: Lovable (lovable.dev) - Announced: 2025-01-24T18:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://lovable.dev/blog/2025-01-24-supercharge-your-ai-projects-lovable-x-replicate-integration - Record: https://forck.live/items/12181-how-to-integrate-replicate-with-lovable - Subject: Lovable / AI app builder Lovable's guide explains how developers can integrate Replicate's AI models into Lovable-built applications, covering image generation, text-to-speech, and video generation. It walks through the workflow of using Replicate's Playground and API within Lovable, and lists use cases such as education apps, marketing campaigns, and real-time AI assistants. ### Vercel — Mitigating Denial of Wallet risks with Vercel - Company: Vercel (vercel.com) - Announced: 2025-01-24T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/mitigating-denial-of-wallet-risks-with-vercel - Record: https://forck.live/items/20905-mitigating-denial-of-wallet-risks-with-vercel - Subject: v0 / AI SDK / AI Gateway Vercel published a guide explaining Denial of Wallet (DoW) attacks, which exploit cloud scalability to drive up operational costs, and describes its built-in defenses including rate limiting, Web Application Firewall, spend controls, and bot protection to help customers mitigate such attacks. ### LG AI Research — [NeurIPS 2024 Series] Attention-Based Data Augmentation Techniques to Enhance Performance of Tabular Data Prediction - Company: LG AI Research (lgresearch.ai) - Announced: 2025-01-24 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.lgresearch.ai/blog/view?seq=520 - Record: https://forck.live/items/4810-neurips-2024-series-attention-based-data-augmentation-techniques-to-enhance - Subject: EXAONE LG AI Research presented their paper 'AGATa: Attention-Guided Augmentation for Tabular Data in Contrastive Learning' at the NeurIPS 2024 TRL Workshop, introducing a novel attention-guided data augmentation technique to improve contrastive learning performance on tabular data while preserving its structural features. ### Replicate — You can now fine-tune open-source video models - Company: Replicate (replicate.com) - Announced: 2025-01-24 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replicate.com/blog/fine-tune-video - Record: https://forck.live/items/2503-you-can-now-fine-tune-open-source-video-models - Subject: Platform Replicate announces support for fine-tuning open-source video models, specifically HunyuanVideo, using an adapted version of Musubi Tuner. The post provides a step-by-step guide to create training data, train a LoRA model, and generate videos. ### Hugging Face — We now support VLMs in smolagents! - Company: Hugging Face (huggingface.co) - Announced: 2025-01-24 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/smolagents-can-see - Record: https://forck.live/items/1763-we-now-support-vlms-in-smolagents - Subject: Platform smolagents now supports vision language models (VLMs) by allowing images to be passed to the agent at start or dynamically via callbacks, enabling use cases like autonomous web browsing. ### OpenAI — Operator System Card - Company: OpenAI (openai.com) - Announced: 2025-01-23T10:00:00+00:00 - Category: safety-policy-update - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/operator-system-card - Record: https://forck.live/items/647-operator-system-card - Subject: GPT / ChatGPT / API OpenAI published a system card for Operator that describes safety mitigations against prompt injection and jailbreaks, privacy protections, external red teaming, and ongoing safety work. ### NVIDIA — Mastering Long Contexts in LLMs with KVPress - Company: NVIDIA (nvidia.com) - Announced: 2025-01-23T08:03:03+00:00 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/nvidia/kvpress - Record: https://forck.live/items/1764-mastering-long-contexts-in-llms-with-kvpress NVIDIA announces KVPress, a Python toolkit for compressing KV cache in LLMs to enable memory-efficient long-context processing. ### Upstage — Turn Charts into LLM-Actionable Data: Introducing Chart Recognition in Upstage Document Parse - Company: Upstage (upstage.ai) - Announced: 2025-01-23 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.upstage.ai/blog/en/introducing-chart-recognition-in-upstage-document-parse - Record: https://forck.live/items/18475-turn-charts-into-llm-actionable-data-introducing-chart-recognition-in-upstage - Subject: Solar Pro / Solar Mini Upstage added chart recognition to Document Parse, enabling the parser to convert chart images into structured HTML. Internal benchmarks show a 6.49% increase in accuracy on chart-related document QA tasks and processing times up to 13x faster than LlamaParse. ### LG AI Research — [NeurIPS 2024 Series] B-splines-Based Input Value Normalization Method for Tabular Data - Company: LG AI Research (lgresearch.ai) - Announced: 2025-01-23 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.lgresearch.ai/blog/view?seq=518 - Record: https://forck.live/items/4811-neurips-2024-series-b-splines-based-input-value-normalization-method-for - Subject: EXAONE LG AI Research presented a new normalization technique for tabular data based on B-splines at the NeurIPS 2024 TRL Workshop. The technique dynamically adjusts input normalization to improve neural network learning on tabular data. ### Hugging Face — SmolVLM Grows Smaller – Introducing the 256M & 500M Models! - Company: Hugging Face (huggingface.co) - Announced: 2025-01-23 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://huggingface.co/blog/smolervlm - Record: https://forck.live/items/1765-smolvlm-grows-smaller-introducing-the-256m-500m-models - Subject: Platform Hugging Face announces two new small vision-language models: SmolVLM-256M (256M parameters) and SmolVLM-500M (500M parameters), which are the smallest VLMs in the world, with improved efficiency and performance. ### Lovable — How a Canadian Infanteer Built 10 Apps in Record Time with Lovable.dev - Company: Lovable (lovable.dev) - Announced: 2025-01-22T18:05:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://lovable.dev/blog/2025-01-20-how-a-canadian-infanteer-built-10-apps-in-record-time-with-lovable-dev - Record: https://forck.live/items/12182-how-a-canadian-infanteer-built-10-apps-in-record-time-with-lovable-dev - Subject: Lovable / AI app builder Tomas, a Toronto-based developer and Canadian Infanteer with 15+ years of coding experience, used Lovable to launch 10 apps, including two in active production. His most notable project, Dummy Forms, won third place in a Lovable hackathon and ranked fifth on Product Hunt, attracting 2,100 unique visitors, 584 signups, and over 190 forms created in eight days. He shares best practices such as prioritizing feature planning, leveraging knowledge panels, and integrating Stripe early in development. ### Lovable — How a Venture Capitalist Rebuilt His Website and Internal Data Platform with Lovable - Company: Lovable (lovable.dev) - Announced: 2025-01-22T18:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://lovable.dev/blog/2025-01-23-how-a-venture-capitalist-rebuilt-his-website-and-internal-data-platform-with-lovable - Record: https://forck.live/items/12183-how-a-venture-capitalist-rebuilt-his-website-and-internal-data-platform-with - Subject: Lovable / AI app builder A venture capitalist at Alpine Ventures rebuilt his website in 30 minutes and reconstructed 80% of an internal backend data system using Lovable, an AI development tool. The rebuilt website replaced a basic Squarespace design with a more professional version, and the data platform improvements increased his startup screening process speed by 10x. Lovable handled backend logic, third-party API integrations, frontend design, load balancing, and error handling. ### OpenAI — Bertelsmann powers creativity and productivity with OpenAI - Company: OpenAI (openai.com) - Announced: 2025-01-22T17:00:00+00:00 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/bertelsmann-powers-creativity-and-productivity-with-openai - Record: https://forck.live/items/649-bertelsmann-powers-creativity-and-productivity-with-openai - Subject: GPT / ChatGPT / API Bertelsmann announces integration of OpenAI's technology across its global brands. ### Vercel — Vercel acquires Tremor to invest in open source React components - Company: Vercel (vercel.com) - Announced: 2025-01-22T13:00:00+00:00 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://vercel.com/blog/vercel-acquires-tremor - Record: https://forck.live/items/20906-vercel-acquires-tremor-to-invest-in-open-source-react-components - Subject: v0 / AI SDK / AI Gateway Vercel acquired Tremor, an open source React component library for dashboards and charts. Tremor's cofounders will join Vercel's Design Engineering team to work on the Vercel Dashboard and v0. Tremor's Pro product, Tremor Blocks, will become free and open source under an MIT license. ### Lovable — Reinventing Design Systems - Company: Lovable (lovable.dev) - Announced: 2025-01-22T11:30:00+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://lovable.dev/blog/2025-01-24-reinventing-design-systems - Record: https://forck.live/items/12185-reinventing-design-systems - Subject: Lovable / AI app builder The role of design systems in modern product development is evolving. Traditionally, design systems have been collections of reusable components and stylistic decisions, but with Lovable, the process is being reimagined to fit the needs of builders in an AI-driven world. ### Lovable — Lovable's two Failed Launches and What we got wrong about PLG - Company: Lovable (lovable.dev) - Announced: 2025-01-22T11:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://lovable.dev/blog/2025-01-28-lovables-two-failed-launches-and-what-we-got-wrong-about-plg - Record: https://forck.live/items/12186-lovable-s-two-failed-launches-and-what-we-got-wrong-about-plg - Subject: Lovable / AI app builder Lovable's early launches as GPT Engineer failed due to assumptions over validation, brand disconnect, and timing challenges. The team rebranded to Lovable, integrated with Supabase for backend capabilities, and focused on user needs, distribution, and feedback to achieve product-market fit. ### Lovable — Software Development Lifecycle 2.0 - Company: Lovable (lovable.dev) - Announced: 2025-01-22T10:30:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://lovable.dev/blog/2025-01-21-revolutionizing-the-software-development-lifecycle - Record: https://forck.live/items/12187-software-development-lifecycle-2-0 - Subject: Lovable / AI app builder Lovable is positioning its AI-driven platform as a way to compress the traditional software development lifecycle, allowing a single person to handle prototyping, design, coding, and deployment. The company emphasizes faster iteration and lower barriers for non-developers, while acknowledging challenges around balancing speed and quality and distinguishing live from work-in-progress states. ### OpenAI — Trading inference-time compute for adversarial robustness - Company: OpenAI (openai.com) - Announced: 2025-01-22T10:00:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/trading-inference-time-compute-for-adversarial-robustness - Record: https://forck.live/items/650-trading-inference-time-compute-for-adversarial-robustness - Subject: GPT / ChatGPT / API OpenAI published an article titled 'Trading Inference-Time Compute for Adversarial Robustness'. ### Aleph Alpha — Aleph Alpha Announces Launch of a Pioneering New Tokenizer-free (T-Free) Architecture Innovation for Next-Generation Sovereign LLMs and a New Powerful Collaboration with AMD and Schwarz Digits - Company: Aleph Alpha (aleph-alpha.com) - Announced: 2025-01-22 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://aleph-alpha.com/en/news/t-free-architecture-amd-schwarz-digits - Record: https://forck.live/items/16862-aleph-alpha-announces-launch-of-a-pioneering-new-tokenizer-free-t-free - Subject: Kolibri / PhariaAI Aleph Alpha announced a tokenizer-free (T-Free) LLM architecture that does not require a tokenizer, enabling more effective fine-tuning for diverse languages, alphabets, and specialized industries. The company also announced a collaboration with AMD and Schwarz Digits to combine this architecture with AMD Instinct MI300 accelerators and the sovereign cloud STACKIT. For Finnish, the architecture achieved a 70 percent reduction in training cost and carbon footprint compared to tokenizer-based models. ### Hugging Face — Hugging Face and FriendliAI partner to supercharge model deployment on the Hub - Company: Hugging Face (huggingface.co) - Announced: 2025-01-22 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/friendliai-partnership - Record: https://forck.live/items/1766-hugging-face-and-friendliai-partner-to-supercharge-model-deployment-on-the-hub - Subject: Platform Hugging Face and FriendliAI announced a partnership integrating FriendliAI Endpoints into the Hugging Face Hub as a deployment option, enabling one-click deployment of generative AI models. ### Lovable — From Idea to Reality: How Bilal Built Aneta, an AI-Powered HR Agent, with Lovable - Company: Lovable (lovable.dev) - Announced: 2025-01-21T16:45:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://lovable.dev/blog/2025-01-21-from-idea-to-reality-how-bilal-built-aneta-an-ai-powered-hr-agent-with-lovable - Record: https://forck.live/items/12188-from-idea-to-reality-how-bilal-built-aneta-an-ai-powered-hr-agent-with-lovable - Subject: Lovable / AI app builder Lovable published a customer story about Bilal, a product designer who built Aneta, an AI-powered HR platform for personalized surveys and insights, in one month using Lovable. The platform replaced the need for a large engineering team and was launched on Product Hunt, attracting 2,000 visitors and a pipeline of 30,000 potential users. Bilal is now pivoting Aneta toward recruitment features. ### Lovable — How to Build a Thriving Business with Lovable? - Company: Lovable (lovable.dev) - Announced: 2025-01-21T15:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://lovable.dev/blog/2025-01-21-from-solo-developer-to-1-million-dream-building-a-thriving-business-with-lovable - Record: https://forck.live/items/12189-how-to-build-a-thriving-business-with-lovable - Subject: Lovable / AI app builder Harry, a former front-end engineer, used the AI platform Lovable to build a business that tripled his revenue within a year. By creating overnight MVPs and demonstrating working prototypes, he closed six-figure contracts and scaled from a solo developer to leading a team of three. The article shares his strategies, including starting small, leveraging communities, and investing in polished presentations. ### OpenAI — Stargate Infrastructure - Company: OpenAI (openai.com) - Announced: 2025-01-21T13:30:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/form/stargate-infrastructure - Record: https://forck.live/items/652-stargate-infrastructure - Subject: GPT / ChatGPT / API OpenAI expresses enthusiasm for its vision of AGI infrastructure and seeks to partner with firms across the data center infrastructure landscape, including power, land, construction, and equipment. ### OpenAI — Announcing The Stargate Project - Company: OpenAI (openai.com) - Announced: 2025-01-21T13:30:00+00:00 - Category: funding-company-news - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/announcing-the-stargate-project - Record: https://forck.live/items/651-announcing-the-stargate-project - Subject: GPT / ChatGPT / API OpenAI announces the Stargate Project. ### Lovable — Building AI Conversations: How Sergio is Crafting Tribbai by Using Lovable - Company: Lovable (lovable.dev) - Announced: 2025-01-21 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://lovable.dev/blog/2025-01-21-building-ai-conversations-how-sergio-is-crafting-tribbai-by-using-lovable - Record: https://forck.live/items/12191-building-ai-conversations-how-sergio-is-crafting-tribbai-by-using-lovable - Subject: Lovable / AI app builder Sergio, a former Web3 professional, used Lovable to build Tribbai, a platform that enables multiple AI roles to converse naturally and collaboratively. The platform features multi-role conversations, memory retention, and tagging systems. Sergio has 65 users on a waiting list and plans to launch a beta after perfecting the conversational capabilities. ### Hugging Face — Yay! Organizations can now publish blog Articles - Company: Hugging Face (huggingface.co) - Announced: 2025-01-20T18:58:24+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/huggingface/blog-articles-for-orgs - Record: https://forck.live/items/1767-yay-organizations-can-now-publish-blog-articles - Subject: Platform Hugging Face announced that organizations can now publish blog articles on their platform, requiring an Enterprise Hub subscription. ### Alibaba — Global-batch load balance almost free lunch to improve your MoE LLM training - Company: Alibaba (alibaba.com) - Announced: 2025-01-20T16:00:03+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://qwenlm.github.io/blog/global-load-balance/ - Record: https://forck.live/items/2299-global-batch-load-balance-almost-free-lunch-to-improve-your-moe-llm-training - Subject: Qwen The Mixture-of-Experts (MoE) architecture, which is a technique for scaling model parameters, and describes how an MoE layer consists of a router and a group of experts. ### Vercel — AI SDK 4.1 - Company: Vercel (vercel.com) - Announced: 2025-01-20T13:00:00+00:00 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://vercel.com/blog/ai-sdk-4-1 - Record: https://forck.live/items/20907-ai-sdk-4-1 - Subject: v0 / AI SDK / AI Gateway Vercel released AI SDK 4.1, adding an experimental generateImage function for image generation via a unified API across providers including Replicate, OpenAI, Google Vertex, and Fireworks. The update also introduces stream transformation and smoothing, simplified chat persistence with useChat, and non-blocking data streaming via createDataStreamResponse. ### Lovable — How a Developer Advocate Built Stunning 3D Projects with Lovable—and Won Big - Company: Lovable (lovable.dev) - Announced: 2025-01-20T11:15:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://lovable.dev/blog/2025-01-20-how-a-developer-advocate-built-stunning-3d-projects-with-lovable-dev-and-won-big - Record: https://forck.live/items/12192-how-a-developer-advocate-built-stunning-3d-projects-with-lovable-and-won-big - Subject: Lovable / AI app builder Lovable.dev published a Q&A with Konstantin, a developer advocate from Ukraine, who used the platform to build 3D environments, integrate a local payment system, and win a hackathon. He describes starting with a simple text prompt to generate a 3D space and iteratively adding features like real-time physics and collision detection. The article shares his tips for success, including starting small, testing locally, and using external tools for advanced iterations.