# forck.live — 16–22 December 2024 > 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 2024-W51, 16–22 December 2024, 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/2024-W51 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### OpenAI — Deliberative alignment: reasoning enables safer language models - Company: OpenAI (openai.com) - Announced: 2024-12-20T10:00:00+00:00 - Category: safety-policy-update - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/deliberative-alignment - Record: https://forck.live/items/656-deliberative-alignment-reasoning-enables-safer-language-models - Subject: GPT / ChatGPT / API OpenAI introduces a new alignment strategy called deliberative alignment for o1 models, which teaches safety specifications and reasoning over them. ### Supabase — Supabase Launch Week 13 Hackathon Winners - Company: Supabase (supabase.com) - Announced: 2024-12-20 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://supabase.com/blog/lw13-hackathon-winners - Record: https://forck.live/items/21377-supabase-launch-week-13-hackathon-winners - Subject: Supabase Supabase announced the winners of its Launch Week 13 Hackathon, recognizing projects such as Brainrot GPT, SupaSketch, Stravation, and others across categories including best overall project, best use of AI, most fun, most visually pleasing, and best project built with bolt.new. ### Supabase — Calendars in Postgres using Foreign Data Wrappers - Company: Supabase (supabase.com) - Announced: 2024-12-20 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://supabase.com/blog/calendars-in-postgres-using-foreign-data-wrappers - Record: https://forck.live/items/21376-calendars-in-postgres-using-foreign-data-wrappers - Subject: Supabase Supabase released Foreign Data Wrappers for Cal.com, enabling users to create event bookings directly from Postgres within a single transaction. The wrappers are built with the Wrappers framework and support Wasm for API-based services. ### Supabase — AI Hackathon at Y Combinator - Company: Supabase (supabase.com) - Announced: 2024-12-20 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://supabase.com/blog/ai-hackathon-at-y-combinator - Record: https://forck.live/items/21375-ai-hackathon-at-y-combinator - Subject: Supabase A few weeks ago we hosted an AI hackathon at Y Combinator in San Francisco. Over 150 people flew in from all around the world, including Greece, Australia, Turkey, New Zealand, and South Africa. ### Hugging Face — Evaluating Audio Reasoning with Big Bench Audio - Company: Hugging Face (huggingface.co) - Announced: 2024-12-20 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/big-bench-audio-release - Record: https://forck.live/items/1777-evaluating-audio-reasoning-with-big-bench-audio - Subject: Platform Artificial Analysis releases Big Bench Audio, a new evaluation dataset for assessing audio reasoning, and presents benchmark results for GPT-4o and Gemini 1.5 series models across speech and text modalities, revealing a significant speech reasoning gap. ### Supabase — How to Hack the Base! - Company: Supabase (supabase.com) - Announced: 2024-12-19 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://supabase.com/blog/how-to-hack-the-base - Record: https://forck.live/items/21378-how-to-hack-the-base - Subject: Supabase Supabase published a walkthrough of its first hack-the-base challenge, detailing nine flags and the techniques used to find them, including inspecting HTML, querying DNS TXT records, and bypassing client-side validation. ### Cartesia — State of voice AI 2024 - Company: Cartesia (cartesia.ai) - Announced: 2024-12-19 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.cartesia.ai/blog/state-of-voice-ai-2024 - Record: https://forck.live/items/8025-state-of-voice-ai-2024 - Subject: Sonic / Ink Cartesia published its first State of Voice report for 2024, highlighting key infrastructure breakthroughs and emerging use cases in voice AI, and looking ahead to 2025. ### Hugging Face — Finally, a Replacement for BERT: Introducing ModernBERT - Company: Hugging Face (huggingface.co) - Announced: 2024-12-19 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://huggingface.co/blog/modernbert - Record: https://forck.live/items/1778-finally-a-replacement-for-bert-introducing-modernbert - Subject: Platform Answer.AI and LightOn released ModernBERT, a family of state-of-the-art encoder-only models that improve upon BERT with longer context (8192 tokens), faster processing, and better downstream performance, available in base (149M params) and large (395M params) sizes. ### Vercel — Optimizing secure build infrastructure with Secure Compute - Company: Vercel (vercel.com) - Announced: 2024-12-18T13:00:00+00:00 - Category: infrastructure-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://vercel.com/blog/optimizing-secure-builds-with-hive-and-secure-compute - Record: https://forck.live/items/20911-optimizing-secure-build-infrastructure-with-secure-compute - Subject: v0 / AI SDK / AI Gateway Vercel describes the integration of its internal compute platform Hive with Secure Compute, reducing provisioning times from 90 seconds to 5 seconds and improving build performance by an average of 30% for sensitive workloads. The post explains the key components—Connectors, Key Exchange Service, WireGuard, and Network Routing—that enable secure, private connections without compromising performance. ### Runway — Runway Launches Global Talent Network - Company: Runway (runway.com) - Announced: 2024-12-18T10:10:10.110000+00:00 - Category: product-launch - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://runway.com/news/runway-launches-global-talent-network - Record: https://forck.live/items/7915-runway-launches-global-talent-network - Subject: Gen / Aleph Runway launched the Runway Talent Network, a media platform for creatives to showcase their work and find job opportunities. The network includes independent artists and leading production companies such as EDGLRD, TOOL of North America, and others. The platform is now live at talent.runwayml.com. ### Odyssey — World Models for Film, Gaming, and Beyond - Company: Odyssey (odyssey.systems) - Announced: 2024-12-18 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://odyssey.systems/introducing-explorer - Record: https://forck.live/items/18606-world-models-for-film-gaming-and-beyond - Subject: Odyssey world models Odyssey introduced Explorer, an image-to-world model that generates photorealistic 3D worlds from a single image. The model uses gaussian splats as its 3D representation and can be edited in creative tools. Explorer is still early, with generation times averaging 10 minutes. ### LG AI Research — AI-Powered PCB Design Automation - Company: LG AI Research (lgresearch.ai) - Announced: 2024-12-18 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.lgresearch.ai/blog/view?seq=510 - Record: https://forck.live/items/4815-ai-powered-pcb-design-automation - Subject: EXAONE LG AI Research describes a hybrid approach combining grid-based and continuous-space methods for PCB auto-routing, using a novel Pathfinder+ algorithm to iteratively resolve constraint violations and accelerate exploration. ### Hugging Face — Bamba: Inference-Efficient Hybrid Mamba2 Model - Company: Hugging Face (huggingface.co) - Announced: 2024-12-18 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://huggingface.co/blog/bamba - Record: https://forck.live/items/1779-bamba-inference-efficient-hybrid-mamba2-model - Subject: Platform IBM, Princeton, CMU, and UIUC announce Bamba-9B, a hybrid Mamba2 model trained on 2.2T tokens with open data, demonstrating 2.5x throughput improvement and 2x latency speedup over standard transformers in vLLM, and released with support in transformers, vLLM, TRL, and llama.cpp. ### Vercel — The rise of the AI crawler - Company: Vercel (vercel.com) - Announced: 2024-12-17T13:00:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://vercel.com/blog/the-rise-of-the-ai-crawler - Record: https://forck.live/items/20912-the-rise-of-the-ai-crawler - Subject: v0 / AI SDK / AI Gateway Vercel and MERJ analyzed AI crawler traffic across Vercel's network, finding that GPTBot generated 569 million requests and Claude 370 million in the past month, together about 20% of Googlebot's 4.5 billion requests. None of the major AI crawlers (OpenAI, Anthropic, Meta, ByteDance, Perplexity) currently render JavaScript, while Google's Gemini and AppleBot do. ChatGPT spends 34.82% of its fetches on 404 pages and Claude 34.16%, compared to Googlebot's 8.22%. ### TII — Welcome to the Falcon 3 Family of Open Models! - Company: TII (tii.ae) - Announced: 2024-12-17T12:00:00+00:00 - Category: open-weight-release - Coverage: not counted - Announcement: yes - Group: models - Source: https://falcon-lm.github.io/blog/falcon-3/ - Record: https://forck.live/items/18417-welcome-to-the-falcon-3-family-of-open-models - Subject: Falcon LLM TII released Falcon3, a family of five decoder-only open-weight LLMs under 10B parameters. The models include transformer-based variants (1B, 3B, 7B, 10B) and a Mamba-based 7B variant, all supporting up to 32k context length (1B: 8k). Falcon3-10B-Base achieves state-of-the-art zero-shot and few-shot performance among models under 13B parameters, and the instruct versions outperform all instruct models under 13B on the open leaderboard. ### Hugging Face — Benchmarking Language Model Performance on 5th Gen Xeon at GCP - Company: Hugging Face (huggingface.co) - Announced: 2024-12-17 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/intel-gcp-c4 - Record: https://forck.live/items/1781-benchmarking-language-model-performance-on-5th-gen-xeon-at-gcp - Subject: Platform The post benchmarks text embedding and text generation on Google Cloud C4 (5th Gen Xeon) vs N2 (3rd Gen Xeon), showing C4 has 10x-24x higher throughput for embedding and 2.3x-3.6x for generation, with TCO advantages. ### Hugging Face — Welcome to the Falcon 3 Family of Open Models! - Company: Hugging Face (huggingface.co) - Announced: 2024-12-17 - Category: open-weight-release - Coverage: not counted - Announcement: yes - Group: models - Source: https://huggingface.co/blog/falcon3 - Record: https://forck.live/items/1780-welcome-to-the-falcon-3-family-of-open-models - Subject: Platform The Technology Innovation Institute (TII) released the Falcon3 family of five open decoder-only large language models under 10 billion parameters: Falcon3-1B-Base, Falcon3-3B-Base, Falcon3-Mamba-7B-Base, Falcon3-7B-Base, and Falcon3-10B-Base. The models emphasize science, math, and code capabilities, with the 7B model pretrained on 14 trillion tokens using 1024 H100 GPUs and the 10B model created via depth up-scaling. Falcon3-10B-Base achieves state-of-the-art zero-shot and few-shot performance for models under 13B parameters, and all transformer-based models are compatible with Llama architecture. ### OpenAI — OpenAI o1 and new tools for developers - Company: OpenAI (openai.com) - Announced: 2024-12-17 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://openai.com/index/o1-and-new-tools-for-developers - Record: https://forck.live/items/657-openai-o1-and-new-tools-for-developers - Subject: GPT / ChatGPT / API OpenAI introduces the o1 model, Realtime API improvements, and a new fine-tuning method for developers. ### Runway — NYU Tisch Graduate Program brings Runway tools to the Martin Scorsese Virtual Production Center - Company: Runway (runway.com) - Announced: 2024-12-16T10:10:10.110000+00:00 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://runway.com/news/nyu-tisch-runway-scorsese-virtual-production-center - Record: https://forck.live/items/7916-nyu-tisch-graduate-program-brings-runway-tools-to-the-martin-scorsese-virtual - Subject: Gen / Aleph Runway's AI video tools are being incorporated into NYU Tisch School of the Arts' Masters of Professional Studies in Virtual Production program for a new spring 2025 course at the Martin Scorsese Virtual Production Center. ### Replicate — AI video is having its Stable Diffusion moment - Company: Replicate (replicate.com) - Announced: 2024-12-16 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://replicate.com/blog/ai-video-is-having-its-stable-diffusion-moment - Record: https://forck.live/items/2505-ai-video-is-having-its-stable-diffusion-moment - Subject: Platform The surge in AI video models following OpenAI's Sora announcement, comparing several models available on Replicate, including their quality, speed, and open-source status. ### Hugging Face — Introducing the Synthetic Data Generator - Build Datasets with Natural Language - Company: Hugging Face (huggingface.co) - Announced: 2024-12-16 - Category: product-launch - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/synthetic-data-generator - Record: https://forck.live/items/1782-introducing-the-synthetic-data-generator-build-datasets-with-natural-language - Subject: Platform Hugging Face introduces the Synthetic Data Generator, a no-code application for creating custom datasets using LLMs. It supports text classification and chat datasets, integrates with Argilla for review and AutoTrain for model training, and is powered by the free Hugging Face API.