AI

AI Research & Development

Ma Jiaqi taught MiniMax engineers a hard lesson about forgotten tokens

MiniMax's internal investigation into why its M2 model couldn't output the name 'Ma Jiaqi' revealed a structural mismatch between pre-training vocabulary and post-training data distribution. The root cause: low-frequency tokens' lm_head vectors drift during SFT, losing generation ability while retaining understanding. A full-vocabulary coverage fix resolved the issue and also mitigated language mixing in Japanese.

2026-07-04

Research analysis

AI document corruption in delegated workflows: what a new stress test reveals

The DELEGATE-52 benchmark evaluates AI systems on long-horizon delegated document editing tasks, finding that frontier models accumulate semantic fidelity loss of 19–34% over 20 iterations. Python workflows showed less than 1% degradation on average, but the study underscores that reliable long-horizon delegation remains an open challenge.

2026-07-04

LLM Inference Optimization

Aleph Alpha builds a theoretical inference model to decode DeepSeek V3 performance from hardware primitives

Aleph Alpha's theoretical model predicts DeepSeek V3 inference performance from hardware parameters alone, revealing how GPU count and interconnect bandwidth shift the bottleneck between compute, memory, and communication.

2026-07-04

Cryptographic Privacy

Vega brings zero-knowledge proofs to identity verification in under 100 ms

Vega enables users to prove facts from government IDs without revealing the credential itself, using ZKPs generated in under 100 ms on commodity hardware. It targets real-world formats like mobile driver's licenses and the EU Digital Identity Wallet, with a fold-and-reuse scheme that reduces proof cost for repeated presentations.

2026-07-04

AI Hardware

Spacex is developing an ai device prototype, report says, but musk denies it

The Wall Street Journal reports that SpaceX has prototyped a sleek AI device smaller than an iPhone, integrating xAI technology and a proprietary OS. Musk calls the report 'utterly false,' while analysts note SpaceX's manufacturing and wireless ambitions could make such a move plausible.

2026-07-04

Artificial Intelligence

Microsoft open-sources Data Formulator 0.7 for enterprise AI analytics

Data Formulator 0.7 combines data connectors, context-aware agents, and an iterative workspace to let enterprise teams prepare, explore, and visualize data across fragmented systems without coding. The open-source release aims to reduce integration work and make analytical workflows reproducible.

2026-07-04

Artificial Intelligence

Microsoft's Phi-4 Model Redefines Efficiency in Breakthrough Research

Microsoft's Phi-4 model achieves state-of-the-art efficiency, matching larger models in reasoning tasks with significantly fewer parameters. Published on May 15, 2025, the research paper reexamines assumptions about scaling laws in AI.

2026-07-03

OpenAI launches Prism: a free LaTeX workspace powered by GPT-5.2 for scientists

Avec Prism, OpenAI propose un environnement de rédaction scientifique LaTeX gratuit, intégrant GPT-5.2 pour la relecture, les citations et la mise en forme automatique. L'outil, collaboratif et sans frais, pourrait bien changer la donne pour les chercheurs.

2026-07-03

Genomic Reanalysis

Talos, an open-source tool, automates reanalysis of genomic data to speed rare disease diagnosis

Talos is an open-source tool that automates genomic reanalysis for rare disease. Tested on nearly 5,000 patients, it delivered 241 new diagnoses within weeks of new evidence emerging, with a low false-positive rate that makes frequent reanalysis sustainable.

2026-07-03

Artificial Intelligence

Microsoft releases gridsfm, a lightweight foundation model for power grid optimization

Microsoft's GridSFM is a neural network that solves AC optimal power flow (AC-OPF) in milliseconds across grids of up to 80,000 buses, offering a fast, accurate approximation that can serve standalone or as a warm-start for traditional solvers. The open-source model aims to transform grid operations from reactive to proactive optimization.

2026-07-03

Aleph Alpha commentary

The case against enshittification: why specialized, sovereign ai beats generic pilots every time

Aleph Alpha warns that generic AI pilots are undermining enterprise trust through a pattern of overpromise and underdeliver, a phenomenon dubbed 'enshittification.' The company advocates for sovereign, domain-specific AI agents built through close customer co-creation.

2026-07-03

Tokenizer-Free Architecture

Aleph Alpha unveils T-Free: a tokenizer-free architecture for sovereign AI

Aleph Alpha unveils T-Free, a tokenizer-free LLM architecture that maps words directly to vectors. The approach delivers nearly seven characters per vector versus the typical four, cutting costs and energy use while improving performance on specialized domains and low-resource languages.

2026-07-03