# forck.live — 31 July – 6 August 2023 > 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 2023-W31, 31 July – 6 August 2023, 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/2023-W31 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### Replit — Fewer Restarts and Faster Networking for All - Company: Replit (replit.com) - Announced: 2023-08-04 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/faster-networking - Record: https://forck.live/items/18074-fewer-restarts-and-faster-networking-for-all - Subject: Replit Agent Replit is expanding a VM upgrade that reduces container restarts to all users, including free-tier users, and is upgrading all network connections to Google Cloud's Premium Tier, which is 40%+ faster than Standard tier. The changes are enabled by Replit's Margarine storage system and a deeper partnership with Google. ### LG AI Research — [CVPR 2023] Looking at CV Research Trends and the Future of Image Captioning at CVPR 2023 - Company: LG AI Research (lgresearch.ai) - Announced: 2023-08-04 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.lgresearch.ai/blog/view?seq=335 - Record: https://forck.live/items/4895-cvpr-2023-looking-at-cv-research-trends-and-the-future-of-image-captioning-at - Subject: EXAONE LG AI Research discusses computer vision research trends at CVPR 2023 and reviews the NICE workshop on zero-shot image captioning evaluation. ### Hugging Face — Deploy MusicGen in no time with Inference Endpoints - Company: Hugging Face (huggingface.co) - Announced: 2023-08-04 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/run-musicgen-as-an-api - Record: https://forck.live/items/2015-deploy-musicgen-in-no-time-with-inference-endpoints - Subject: Platform To deploy the MusicGen model using Hugging Face Inference Endpoints with a custom handler, enabling music generation from text prompts via an API. ### Harvey — Announcing Harvey's Security Advisory Board - Company: Harvey (harvey.ai) - Announced: 2023-08-02T21:36:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.harvey.ai/blog/announcing-harvey-s-security-advisory-board - Record: https://forck.live/items/17737-announcing-harvey-s-security-advisory-board - Subject: Harvey Harvey announced the formation of a Security Advisory Board composed of information security representatives from institutions it engages with and the infrastructure it operates on. The board will advise Harvey's leadership on defining security best practices for developing generative AI systems that process highly-sensitive data. ### Hugging Face — Towards Encrypted Large Language Models with FHE - Company: Hugging Face (huggingface.co) - Announced: 2023-08-02 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/encrypted-llm - Record: https://forck.live/items/2017-towards-encrypted-large-language-models-with-fhe - Subject: Platform Zama demonstrates how to use Fully Homomorphic Encryption (FHE) to run parts of a GPT2 model on encrypted data, protecting both user privacy and model intellectual property, by adapting the Hugging Face transformers library with quantized operators. ### Hugging Face — Huggy Lingo: Using Machine Learning to Improve Language Metadata on the Hugging Face Hub - Company: Hugging Face (huggingface.co) - Announced: 2023-08-02 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/huggy-lingo - Record: https://forck.live/items/2016-huggy-lingo-using-machine-learning-to-improve-language-metadata-on-the-hugging - Subject: Platform Hugging Face is using the fastText language identification model to automatically detect and add language metadata to datasets on the Hub that lack it, improving discoverability. ### Upstage — Upstage's 70B Language Model Outperforms GPT-3.5, Becomes Global No.1 - Company: Upstage (upstage.ai) - Announced: 2023-08-01 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://www.upstage.ai/news/upstage-huggingface-llm-no1 - Record: https://forck.live/items/18451-upstage-s-70b-language-model-outperforms-gpt-3-5-becomes-global-no-1 - Subject: Solar Pro / Solar Mini Upstage released a 70-billion-parameter language model that achieved an average score of 72.3 on the HuggingFace Open LLM Leaderboard, surpassing GPT-3.5's score of 71.9 and claiming the top spot. The model was fine-tuned with LLaMA 2 and trained with 70 billion parameters, outperforming the previous leader, Stability AI's Stable Beluga 2 (71.4). This marks the first time an open-source model has outperformed a global tech giant's model on this benchmark. ### Replit — Rebuilding the Replit Filetree for Superior Performance and Accessibility - Company: Replit (replit.com) - Announced: 2023-08-01 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/filetree-updates - Record: https://forck.live/items/18101-rebuilding-the-replit-filetree-for-superior-performance-and-accessibility - Subject: Replit Agent Replit rebuilt its filetree to improve performance and accessibility. The new filetree uses virtualization to render only visible rows, reducing click handler execution time from 344ms to 16ms when expanding a directory with 500 files. It also adds keyboard navigation and ARIA attributes for assistive technologies. ### LG AI Research — [ICASSP 2023]Multi-resolution Sequence Aggregation and Model-Agnostic Framework for Time-Series Forecasting - Company: LG AI Research (lgresearch.ai) - Announced: 2023-08-01 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.lgresearch.ai/blog/view?seq=334 - Record: https://forck.live/items/4896-icassp-2023-multi-resolution-sequence-aggregation-and-model-agnostic-framework - Subject: EXAONE This blog post from LG AI Research describes a research paper presented at ICASSP 2023 that introduces the MAMA framework, a multi-resolution sequence aggregation and model-agnostic framework for time-series forecasting. ### Hugging Face — Open-sourcing Knowledge Distillation Code and Weights of SD-Small and SD-Tiny - Company: Hugging Face (huggingface.co) - Announced: 2023-08-01 - Category: open-weight-release - Coverage: not counted - Announcement: yes - Group: models - Source: https://huggingface.co/blog/sd_distillation - Record: https://forck.live/items/2019-open-sourcing-knowledge-distillation-code-and-weights-of-sd-small-and-sd-tiny - Subject: Platform Segmind open-sources the weights and training code of two compressed Stable Diffusion models, SD-Small and SD-Tiny, which have 35% and 55% fewer parameters respectively than the base model, trained using knowledge distillation. ### Hugging Face — Practical 3D Asset Generation: A Step-by-Step Guide - Company: Hugging Face (huggingface.co) - Announced: 2023-08-01 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/3d-assets - Record: https://forck.live/items/2018-practical-3d-asset-generation-a-step-by-step-guide - Subject: Platform A tutorial on creating low-fidelity 3D assets using generative AI tools such as Shap-E and Dream Textures. ### Replit — Replit Deployments: 3x Faster and Even More Features - Company: Replit (replit.com) - Announced: 2023-07-31 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/deployments-updates - Record: https://forck.live/items/18040-replit-deployments-3x-faster-and-even-more-features - Subject: Replit Agent Replit upgraded its Deployments hosting service with 2-3x faster deployment, fewer failures, and new features like dynamic header buttons, NATS-based logging, and support for the X-Forwarded-For header. The improvements, based on user feedback, include upgraded build machines, removed IOps limits, and fixes for symlink, container permissions, and crash-loop errors.