AI Daily Brief — 2026-10-03

Kolibri is an open-weight large language model (LLM) from Aleph Alpha, released on October 3, 2026, under the Apache 2.0 license. This mixture-of-experts model has 78 billion parameters but utilizes only about 3.5 billion for each token. Trained from scratch on German and Finnish infrastructure, Kolibri's weights are available on Hugging Face. It demonstrates strong performance in Aleph Alpha's evaluations, particularly in German and English tokenization, where it leads other open models of its size.

Kolibri is an open-weight model with two versions: Kolibri Origin and Kolibri. Kolibri Origin, finished pre-training on June 11, 2026, has 30.6B total parameters and 3.27B active parameters per token. Kolibri, finished pre-training on September 11, 2026, has 78.1B total parameters and 3.46B active parameters per token. Both models feature 50 layers, with Kolibri Origin using 2 dense + 48 MoE layers and Kolibri using all MoE layers. Kolibri has a longer pre-training context length of 16,384 (16k) and a longest trained length of 262,144 (256k) compared to Kolibri Origin's 8,192 (8k) and 65,536 (64k) respectively.

DS4, created by the developer of Redis, enables local LLM inference. It offers impressive performance metrics across various hardware configurations, such as the M5 Max and DGX Spark, with context prefill and generation speeds measured in tokens per second. Users can begin with a quickstart guide, review the hardware matrix, and then integrate their editor, agent, or API client with the local server for efficient operation.

Nvidia has announced a new version of its DGX Spark mini PC, featuring 64 GB of unified memory. This model is priced at $4,999, which is $1,000 more than the 128 GB version that was available at launch. The introduction of this 64 GB model is a response to the ongoing memory shortage, allowing the DGX Spark to continue running AI models locally despite supply constraints.

Meta has open-sourced code for its Muse AI gadgets, allowing users to create their own AI-powered devices. The company is also distributing 5,000 Muse Home Link gadgets, which enable AI agents to control smart home devices like lights and TVs using community-built skills. Interested users can join a waitlist for the Muse Home Link, with shipping expected to begin this month. This initiative aims to foster innovation and expand the utility of AI in everyday environments.

AstaBrief, a fast report-generation model in Asta, was open-sourced. For supervised fine-tuning (SFT), 47K usable training examples were generated from filtered queries using the multi-step ScholarQA pipeline. This pipeline retrieved literature, organized material, and synthesized evidence into cited reports using proprietary systems like Claude 3.5 Sonnet, Claude 3.7 Sonnet, o3, o4-mini, and GPT-4.1. The development focused on ensuring scientific faithfulness, addressing issues like meandering reports, incorrect citation support, and overstating claims beyond what the evidence supports.

An open-source Lego AI generator has been developed, allowing users to provide an AI agent with a model idea. The agent then guides the process, building the model and ultimately producing an LDraw LEGO model. This involves the agent producing a "plan.json" which is interpreted by "generator.py" for execution, resulting in a "model.mpd" file. The software is not sponsored, authorized, or endorsed by the LEGO Group.

Offrun is a workspace designed to manage coding agents, allowing users to integrate existing CLIs without new accounts or API keys. It provides features like worktrees for isolated agent operations, peer review where a second agent checks diffs before commitment, and account management to handle API limits by switching logins. Offrun also offers project memory, notifications for agent status, and rich previews of agent output directly within the chat interface.

Opus 5.5 in Claude and Claude Code offers enhanced capabilities for multi-step tasks, outperforming Opus 5. It can work for longer periods with minimal oversight, making it suitable for extensive coding tasks that require carrying changes through large repositories until tests pass. The model clearly communicates its actions and thinks before each reply. Users should be aware of notifications indicating a switch to older models, as the chat will then continue on that model.

A growing number of AI agents can be accessed via text messages, eliminating the need to download additional apps. For example, Ollie, launched in June 2026, offers free and paid plans starting at $25 per month for 150 messages, with a $100-per-month plan providing up to 1,000 messages. Another agent, Stanley, can connect to a user's Instagram account to suggest content, convert voice notes into scripts and captions, and manage a content calendar to automate post scheduling.

OpenAI's DevDay 2026 announcements indicate a strategic shift to transform ChatGPT into a platform for discovering, launching, and utilizing software. This move, highlighted by Sarah Perez/TechCrunch, suggests a potential disruption to the conventional app store model. While the event also showcased agentic assistants known as Dots and new AI models, the integration of software discovery and usage within ChatGPT could redefine how users interact with applications, challenging established distribution channels.

Supabase Inc., a startup commercializing the open-source PostgreSQL database, has successfully raised $150 million in funding. This significant investment round was led by Singapore's GIC sovereign wealth fund. Additionally, Supabase has agreed to acquire Turso, a company known for offering a database specifically optimized for AI agents. The financial terms of the acquisition remain undisclosed.

Leaked Slack messages reveal OpenAI employees' pushback in June against Greg Brockman's connections to Leading the Future. This internal dissent reportedly contributed to Brockman's decision to renege on a $25M donation. The messages surfaced after OpenAI chief strategy officer Jason Kwon previewed a blog post denying company ties, highlighting internal tensions regarding external affiliations and their impact on significant financial commitments.

The surge of AI video content on social media has deceived viewers and creators, prompting questions about the regulation or banning of AI on these platforms. 7NEWS, a trusted news brand in Australia, covers this issue, delivering continuous coverage on topics that matter most to Australians. They broadcast full episodes at 6pm and weekdays at 12pm and 4pm on Channel 7 and 7plus.

David Robinson, an OpenAI safety employee, has resigned, claiming the company's 'culture is broken.' Robinson believes the discussion should extend beyond specific rules or new laws to address the overall culture within AI companies. His essay suggests that OpenAI's cultural issues are reflective of broader problems within Silicon Valley, rather than solely focusing on CEO Sam Altman's loss of trust among former colleagues.

David Robinson, a leader on OpenAI's Safety Systems team and former head of policy planning, resigned from the company last week. An OpenAI spokesperson confirmed Robinson's departure to Business Insider on Friday. His exit marks a notable change within the organization's safety and policy leadership.

David Robinson, former head of safety and policy at OpenAI, argues that Silicon Valley lacks a safety-centric culture. He emphasizes that AI labs must learn from safety approaches in other fields, stating that the era for trial and error in AI safety is over. Robinson warns that the current industry approach to safety will inevitably lead to more failures unless significant changes are implemented.

Google's Project Suncatcher prototype satellite, developed in collaboration with Planet, has successfully launched into orbit. The satellite was part of the Transporter-18 rideshare mission with SpaceX, and Google's team has confirmed contact and normal operation. This launch marks the beginning of in-space experiments, with the insights gained expected to refine future designs as the mission progresses.

The Cambridge Programme on AI Science & Policy (CASP) is an interdisciplinary research programme at the University of Cambridge. CASP focuses on frontier artificial intelligence, aiming to understand its complexities and implications. This initiative brings together various disciplines to explore the scientific and policy aspects of advanced AI.

OpenAI's recent event experienced some technical difficulties, described as "oopsies." The event, titled "OpenAI Dots Failed Live on Stage," suggests that these issues occurred during a live presentation. Further details about the nature of these failures are not provided in the available information.