Latest AI News

Get the fastest, most accurate AI news every morning

Total articles: 11,532|Added todayToday: 60

Category

Search

Today'sSafety & Ethics

Sunak Warns Trust Could Limit AI Adoption

Rishi Sunak called India’s AI debate “incredibly positive” and warned public distrust could impede adoption.
He urged a shift from the race for advanced AI to spreading “everyday AI” across economies and society.
Sunak cited Mindspark’s half a million pupils and doubled learning rate, and compared ChatGPT’s two-month adoption milestone with older technologies.

FirstpostRead more · 2 min →

Relace Raises $1.6m for AI Compliance

London fintech Relace raises $1.6m (£1.2m) in pre-seed funding to help firms navigate financial regulation with AI
Boost Capital leads the round; Relace selected among 21 companies from 199 applicants for the FCA’s Supercharged Sandbox
Platform flags regulatory gaps; a customer identified 21 previously unknown risks, while firms retain compliance responsibility

cityam.comRead more · 1 min →

DOSS Unveils Operations Platform Suite

DOSS announced Dynamic Apps, Product Suite 2.0 and an evolution of its Dossbot copilot after four years of development.
DOSS cites 55–75% traditional software implementation failures; customers reported 95% faster invoicing and 92% shorter order processing.
DOSS says its implementation service takes four to six months, compared with a typical 12–16 months.

SalesTech StarRead more · 2 min →

FinTurk Partners With Fiduciary Check

FinTurk and Fiduciary Check partnered to offer the CRM to network members at discounts of up to 44%.
Fiduciary Check subsidizes platform costs, and preferred pricing varies by membership tier.
FinTurk’s Finn assistant supports advisor tasks while leaving every action under the advisor’s control.

SalesTech StarRead more · 1 min →

Rimini Street CEO Proposes an “AI Last” ERP Strategy

Rimini Street CEO Seth Ravin says companies should identify business problems and expected ROI before adopting AI in ERP initiatives.
The proposal adds an agentic AI layer over existing ERP systems, with ServiceNow, Palantir, or Microsoft as possible tools.
Moving from SAP ECC 6.0 to SAP S/4HANA can take hundreds of millions of dollars and 3–5 years, Ravin says.

CIO KoreaRead more · 2 min →

Pricefx and Databricks Partner on Pricing AI

Pricefx and Databricks plan to integrate enterprise pricing tools with the Databricks Data + AI Platform.
Three milestones include scheduled data connections, zero-copy lakehouse access, and pricing agents on Databricks Marketplace.
A free notebook will test margin leakage on demo chemicals data in about 15 minutes; further agents are planned in coming months.

SalesTech StarRead more · 2 min →

NVIDIA Releases Nemotron 3.5 Models

NVIDIA released Nemotron 3.5 Super VL and Bolt 74B-A9B on October 15, 2026
Super VL scored 26.5 on Artificial Analysis Intelligence Index and used 46k output tokens per task
Bolt scored 21.4, while Super VL reached about 630 tokens per second on 100k-token prompts

Artificial AnalysisRead more · 1 min →

Trending Keywords

Today'sSafety & Ethics

AMA Sets Governance Framework for Healthcare AI

AMA releases four-question framework for clinical AI agents in healthcare
Survey: 28% of healthcare organizations deployed agentic AI; 44% reported pilots
72% report occasional deployment without IT approval; 47 states introduced over 250 healthcare AI bills in 2025

Healthcare DiveRead more · 1 min →

Cohere Pitches Controlled AI for Utilities

Cohere says its North platform can run within utilities’ secure environments and connect AI to internal knowledge.
OpenAI reported July cybersecurity evaluations circumvented isolation controls; it said utility grid systems were not shown compromised.
Cohere recommends five checks for AI proposals, covering measurable value, agent authority, data flows, validation, and post-pilot licensing.

CohereRead more · 2 min →

Cohere Proposes Competition to Manage Agent Sprawl

Cohere proposes evidence-based competition between enterprise AI agents to manage proliferation without eliminating useful duplication.
A bank example tests incumbent and challenger agents for 30 days across four measures, including token cost per memo.
The challenger may replace the incumbent, be retired, or remain alongside it for different lending tasks, depending on results.

CohereRead more · 2 min →

LangChain Announces October Product Updates

LangChain announced LangSmith and LangChain product updates at Interrupt NYC and London in its October 2026 newsletter.
Managed Deep Agents v0.9 adds contextual agents with Slack, memory, tool use, scheduled tasks and built-in web search.
Newsletter lists workshops, events and a Zip customer account alongside product updates and production agent guidance.

LangChainRead more · 1 min →

NVIDIA Nemotron 3.5 Super VL Launches on Baseten

NVIDIA Nemotron 3.5 Super VL becomes available on Baseten for inference and customization
The open 120B-A12B model handles text, images, video and agent tool use
NVIDIA reports higher scores than Nemotron 3 Super on GDPval, SciCode and GPQA-diamond

BasetenRead more · 1 min →

Meta Releases Muse Spark Open Weights

Meta released Muse Spark Open weights, and Baseten offers the 816B-parameter model through APIs and dedicated deployments.
The model has 42.7B active parameters, a 1M-token context window, and multimodal input for coding-agent workflows.
Hybrid attention lowers KV-cache use from 31 GiB to 8.4 GiB per sequence at 128K context, Baseten says.

BasetenRead more · 2 min →

SGLang Adds Day-0 Muse Spark Open Support

SGLang and Miles announced day-0 serving and training support for Muse Spark Open on October 15, 2026
On 8 NVIDIA B300 GPUs, an OpenHands replay reached 14.8K and 80.3K total tokens per second per GPU
Sequence parallelism with MegaMoE cut p90 time to first token by 28.3–41.9% versus TP4 across seven concurrency levels

LMSYSRead more · 2 min →

Mistral Releases Timestral Forecasting Models

Mistral released Timestral, its first time-series foundation model family, on October 15, 2026.
Its 209M-parameter model ranks first on five metrics across GIFT-Eval, TIME and fev-bench, Mistral says.
Four model sizes run on CPU; the 5M model forecasts one day ahead in 10 ms using 8 threads.

Mistral AIRead more · 3 min →

Meta’s Muse Spark Open Previews on GroqCloud

Meta’s Muse Spark Open enters limited preview on GroqCloud on launch day
The model has 816B total parameters, 42B active per token and 256 experts
GroqCloud reports 350+ tokens per second per user and a 131,072-token context window

GroqRead more · 1 min →

ActionOPD Distills Agent Decisions

ActionOPD trains smaller language-model agents to align their tool-use decisions with teacher models
Training Qwen3-4B and Qwen3-8B produced average gains of 3.3–6.9 percentage points across four benchmarks
The method avoids requirements for shared tokenizers, teacher token probabilities and copied teacher reasoning

HuggingFaceRead more · 1 min →

Cortex Turns Context Gains Into Model Updates

ByteDance Seed researchers introduce Cortex to preserve context-driven gains after context is removed
A five-task-family testbed measures retention across single, batch and sequential model updates
After 50 edits, Cortex-Structural recovered 88.9% of gains and retained 99.8% of non-target performance

HuggingFaceRead more · 1 min →

Meta Releases Muse Spark Open Weights

Meta releases Muse Spark Open weights, calling it the most capable US open-weight model to date
The multimodal model has 816B total parameters, 42B active, and a 1M-token context window
Meta reports up to 3.95× kernel speedups and a 3% end-to-end speedup on AMD MI350

Meta Superintelligence LabsRead more · 1 min →

CriticOPD Improves Math Model Training

CriticOPD uses teacher critiques to improve on-policy distillation for mathematical reasoning
Across six benchmarks, average accuracy rose 4.57 points on Qwen3-1.7B and 4.06 on Qwen3-0.6B
Students rewrite flawed solutions with critique, then train on revisions without seeing the feedback

HuggingFaceRead more · 1 min →

Trending Keywords

Last 7 Days