AI
skim
IQ
Home
Archive
About
☾
EN ▾
← Read the full brief
Podcast
Thursday, 2 April 2026
Duration: 5:17
Listen to the Czech audio edition and follow the transcript below.
▶
0:00
0.75×
1×
1.25×
1.5×
2×
--:--
Transcript
0:00
Welcome to AIskimIQ, your quick daily dive into the world of artificial intelligence. I'm your host, and today is Thursday, April 2nd, 2026. We're covering everything from breakthrough model compression and AI agent security concerns to record-breaking venture funding and chip market shifts.
0:17
Starting with large language models, where we're seeing major compression breakthroughs and some security headaches.
0:23
Caltech researchers claim their PrismML technology achieves radical compression of high-fidelity AI models down to just 1-bit without sacrificing performance, drastically lowering energy consumption.
0:32
Anthropic suffered a second major security breach when hundreds of thousands of lines of code from its AI coding tool leaked, exposing researchers to details about upcoming models and internal architecture.
0:45
New research explores Large Action Models—LLMs specifically trained to conduct data-driven actions—distinguishing them from traditional language models and showing how they're reshaping the AI landscape.
0:55
AI agents are becoming increasingly capable, but that power comes with growing legal and security challenges.
1:01
Anthropic announced that Claude can now use your computer to finish tasks independently, joining competitors in ramping up AI agent capabilities after earlier viral breakthroughs.
1:11
Amazon filed a lawsuit against Perplexity over AI agent liability issues, raising critical questions about who's responsible when autonomous agents make decisions.
1:20
A new guide highlights two major risks of AI agents built with frameworks like OpenClaw: they have access to all your data and passwords, and they make autonomous decisions without constant oversight.
1:33
On the safety front, we're seeing unusual political alliances forming and important lessons from recent breaches.
1:39
An unlikely coalition of billionaire-backed Silicon Valley figures and populists like Bernie Sanders are joining forces to prevent what they see as an AI disaster, though their shared goals remain fragile.
1:52
Anthropic attributed the Claude code leak to human error rather than a security breach, forcing a reckoning for the company that had been preaching trust and safety in AI.
2:03
Researchers Hasan and Dhrubo have developed interpretable AI that significantly boosts cardiovascular disease diagnosis accuracy, demonstrating how explainable AI can improve critical healthcare outcomes.
2:13
Microsoft's Copilot suite continues to expand with new features and enterprise capabilities.
2:18
Microsoft Copilot is using a surprising new technique to generate more accurate and trustworthy research reports, offering users better reliability in AI-driven analysis.
2:27
Microsoft is helping users build websites entirely with AI, offering step-by-step guidance and clear prompts to take projects from concept to launch.
2:36
Microsoft has launched Copilot Cowork in Frontier, though the company still needs to implement specific improvements to enable true enterprise-grade collaboration at scale.
2:45
The video and image generation space is heating up as costs drop and features expand.
2:51
Google's Veo 3.1 Lite slashes API costs in half, targeting developers frustrated by expensive generation—arriving just as OpenAI pulled the plug on its Sora service.
3:01
DomoAI updated its platform with enhanced Talking Avatar features that now easily generate audio files from text, making video creation more accessible.
3:10
PixVerse's new V6 AI video generation tool enables creators to produce multi-shot, cinematic content with automatically synchronized audio, simplifying complex video production workflows.
3:19
NVIDIA and major robotics companies are moving AI from the lab to the factory floor and operating room.
3:26
NVIDIA and robotics companies are testing a shared AI software stack across industrial robots and surgical systems in what researchers call one of the first coordinated deployments of this scale.
3:38
AI is accelerating discovery in unexpected ways, from materials science to molecular simulation.
3:43
Researchers at KIT analyzed materials science literature using LLMs and machine learning, revealing emerging research trends and directions for future investigation.
3:52
Scientists at the University of Manchester created a physics-informed machine learning model that runs stable molecular simulations at extreme temperatures, achieving unprecedented accuracy and speed.
4:02
Venture capital is pouring money into AI at historic levels as the first quarter of 2026 sets new records.
4:09
Q1 2026 shattered venture funding records with $300 billion invested in startups, driven by unprecedented spending on AI compute and frontier model development.
4:18
Venture capital hit $297 billion across 6,000 startups in one quarter, with AI capturing a massive 81% of all funding—marking an historic moment for the industry.
4:29
On the hardware side, NVIDIA faces new competition while AI chip design itself is getting a makeover.
4:36
NVIDIA announced new MLPerf inference records by combining co-designed hardware, software, and models, demonstrating that optimization across all layers drives superior AI factory throughput and lower token costs.
4:47
Chinese chipmakers captured nearly 41% of China's AI accelerator server market last year, significantly eroding NVIDIA's once-dominant position in a critical region.
4:56
Cognichip raised $60 million to develop AI systems that design AI chips themselves, claiming they can cut chip development costs by over 75% and slash timelines in half.
5:07
That's today's AI news roundup on AIskimIQ. We'll be back tomorrow with more stories from the cutting edge of artificial intelligence. Thanks for listening!
← Previous
AIskimIQ Podcast — 2026-04-02 | AIskimIQ