Lectura

Next-generation large language models have left only one of 100 research-level mathematics questions unsolved in the "Leipzig Benchmark."

The dataset was compiled by 49 mathematicians between April and May 2026 to test AI capabilities on high-difficulty problems with known answers. Earlier evaluation stages showed 41 questions completely unsolved in initial attempts, dropping to 2 after multi-run evaluations and heavy-thinking model interventions.

This update (Stage 4) introduced configurations of next-generation models equipped with web search and code execution. The results indicate that the vast majority of questions are now conquered, suggesting that current frontier models approach human-expert levels in specific structured mathematical reasoning tasks.

Note that these are author-reported results on a small sample size (100 questions), and independent third-party reproduction has not yet been observed.

The BuzzASR model suite outperforms Whisper-large-v3 in 77 of 102 languages, reducing the average Character Error Rate (CER) by more than 2.8 times.

Mainstream end-to-end speech recognition models are typically trained on multiple languages, which often leads to poor performance in low-resource languages with limited training data. While monolingual fine-tuning is known to be effective, it had previously been applied only to a small number of languages.

This study scales the monolingual fine-tuning strategy to 102 languages and introduces tokenizer replacement, improving compression rates by an average of 3.3 times. All models, code, and detailed results have been open-sourced.

Large language models fail to outperform traditional non-LLM baselines when simulating human survey responses.

The study by Zihan Chen et al. tested four models across U.S. social attitudes (GSS) and cross-cultural values (WVS). It found that models systematically overestimate the predictive power of demographics, inflating between-segment gaps by two to four times in targeting tasks.

This suggests teams using LLM-generated 'synthetic users' for market or policy decisions may target the wrong segments. The findings come from a preprint and have not yet been peer-reviewed or independently reproduced.

Italian Prime Minister Giorgia Meloni filed an application with the European Union Intellectual Property Office (EUIPO) on October 5 to register her voice as a trademark, aiming to prevent AI-generated deepfakes.

The application includes a four-second recording where she states, "I am Giorgia Meloni." This move follows years of manipulated images and videos of her circulating online, some mistaken for real content.

The application is currently under review. While Italian media note that a trademark alone cannot fully stop others from creating AI audio using her voice, it adds legal hurdles for those attempting to replicate her speech.

According to a Bloomberg report on October 6, Kuaishou's subsidiary Kling AI has selected China International Capital Corporation (CICC), Goldman Sachs, and UBS as underwriters for its planned Hong Kong IPO.

The offering aims to raise at least $1 billion, with a target listing date as early as next year. This follows a $2.8 billion funding round completed in July, which valued the company at approximately $15 billion pre-money.

Sources noted that discussions are ongoing, and details such as the final fundraising size and timeline remain subject to change.

Apple has accused OpenAI and its former employees of violating court rules in their ongoing trade secret lawsuit, claiming defendants used an opposition filing to effectively submit a rebuttal.

According to IT Home, Apple filed a response on October 6 stating that the defendants' opposition document was nine pages long, exceeding the five-page limit, and included new testimony from Chang Liu regarding data erasure. This allegedly violates Rule 7-3(d)(1), which prohibits further debating the motion itself in such filings.

The core issue is Apple's request for a preliminary injunction to prevent its trade secrets from being integrated into OpenAI's hardware development. Oral arguments on this motion are scheduled for October 14.

Anthropic announced a major architectural shift for Claude Cowork: moving tool execution from local user VMs to cloud-based sandboxes.

Previous versions ran inference in the cloud but required a local VM for operations, causing high disk usage, battery drain, and task interruption when laptops closed. The new architecture assigns each session an isolated cloud sandbox, with the desktop app handling only specific file access requests.

This change aims to unlock mobile potential, allowing users to run complex tasks on phones without local compute constraints. According to Anthropic engineer Felix Rieseberg, this directly addresses core user complaints regarding battery life and portability.

McDonald's is facing a proposed nationwide class-action lawsuit in the United States, with plaintiffs alleging that the fast-food giant used an artificial intelligence pricing system to illegally coordinate menu prices.

The suit was filed last Friday (October 2) in federal court in Chicago. The complaint claims McDonald's conspired with independent franchisees to manipulate prices using algorithms trained on non-public transaction data, violating U.S. antitrust laws. Previous Reuters reporting indicated that McDonald's pricing engine analyzes millions of daily transactions across nearly 14,000 locations.

In a statement released Monday, McDonald's rejected the allegations as subjective speculation without factual basis. The company asserted, "No Big Mac or any other item's price is set by AI," emphasizing that franchisees retain pricing authority and that data analytics tools are widely used across industries for recommendations.

This case follows recent U.S. lawsuits against hotels and apartment rentals over algorithmic price coordination. If courts ultimately rule that algorithmic outputs constitute illegal collusion, it could have significant legal implications for large chain enterprises relying on dynamic pricing.

OpenAI is introducing visual display ads that will appear alongside images generated by ChatGPT.

The feature begins rolling out in the U.S. later this month, targeting only free and low-cost subscription tiers. The company states that ads will be clearly labeled and will not influence the answers provided by ChatGPT.

This move aims to attract online advertisers looking to use visual storytelling to reach potential customers. Eventually, these ads are expected to target ChatGPT’s 1.2 billion weekly users as the rollout expands globally.

German industrial robotics startup RobCo announced the closure of a $40 million funding round, pushing its valuation above $1 billion.

The transaction primarily involved current and former employees selling shares to investors including Sequoia and Lightspeed. This represents a doubling of the company's value compared to January.

RobCo specializes in logistics and manufacturing automation, featuring the Alfie mobile robot and RobVision AI engine. The capital will support CEO Roman Hölzl's move to San Francisco to expand the company's U.S. presence.

Nolla Health's AI system has received approval in Utah to autonomously issue acne prescriptions.

While previous AI medical pilots were limited to medication renewals, this is the first case in the US approved for issuing "initial" prescriptions. The pilot gradually reduces human oversight: the first 100 cases require dual physician review, shifting to sample audits thereafter.

The service costs $4.99/month and is restricted to Utah residents aged 18+ with mild-to-moderate acne. If the AI cannot confidently select a treatment, users are referred to a human physician.