Edition: April 24, 2026

Isomorphic Labs, a DeepMind spinoff, is set to begin human trials of AI-designed drugs, marking a significant milestone in the use of artificial intelligence for drug discovery. The company's technology, built on DeepMind's AlphaFold platform, has already shown promise in predicting protein structures and designing safe and effective drugs. The trials will test the efficacy of these molecules in humans, potentially leading to breakthroughs in the treatment of various diseases.

DeepSeek, a Chinese AI firm, has released a preview of its new flagship model V4, which can process longer prompts and is available as an open-source model, offering a cheaper alternative to comparable models from other companies. V4 marks DeepSeek's most significant release since R1 and is a strong alternative to the latest big AI models. The model's performance rivals that of leading closed-source models at a fraction of the cost.

DeepSeek has unveiled its new V4 models, which promise to rival the performance of top proprietary American large language models while significantly reducing inference costs. The models are designed to be highly efficient and can run on Huawei's Ascend family of AI accelerators. DeepSeek V4 introduces novel architectural changes, including a hybrid attention mechanism and lower precision datatypes, to achieve its efficiency gains.

Weiyao Wang, a former Meta employee, has joined Thinking Machines Lab, a move that comes as the AI startup expands and signs a multibillion-dollar cloud deal with Google. This marks the latest example of a talent grab between the two companies, with TML hiring several former Meta researchers. The startup's headcount now stands at around 140, with a valuation of $12 billion.

ComfyUI, a startup that helps creators control AI-generated media, has raised $30 million in funding at a $500 million valuation. The company's node-based workflow gives creators granular control over image, video, and audio outputs from diffusion models. With over 4 million users, ComfyUI is becoming a necessary tool for technical artists and creatives.

Google plans to invest up to $40 billion in AI firm Anthropic, with $10 billion committed immediately and $30 billion to follow if performance targets are met. The investment will support Anthropic's growing computing needs, particularly for its latest model, Mythos. This deal highlights the intense competition for compute capacity in the AI sector.

Apple's CEO Tim Cook is stepping down in September, with hardware chief John Ternus taking over, while Elon Musk's SpaceX has a $60B option to buy Cursor, revealing his AI strategy post-xAI merger. The transition and deal raise questions about the future of tech and AI. Ternus faces a changing ecosystem with challenges to Apple's App Store and developer relationships.

The US military's use of AI systems, such as the Maven Smart System, has significantly accelerated the targeting process, enabling over 1,000 targets to be struck in the first 24 hours of the assault on Iran. This development is explored in journalist Katrina Manson's new book, Project Maven: A Marine Colonel, His Team, and the Dawn of AI Warfare. The project has its roots in an experiment in applying computer vision to drone footage, which began in 2017.

As AI agents become increasingly autonomous in corporate networks, they require interaction infrastructure to operate efficiently, with startups like Band emerging to address this need with dedicated interaction layers. Without such infrastructure, companies face financial liabilities due to unmanaged automation and integration challenges. Band has secured $17 million in funding to develop its solution.

Apple's new CEO, John Ternus, is expected to launch a revolutionary AI product that makes the technology accessible to the masses, a move that is crucial for the company's future. The product is anticipated to simplify AI and make it user-friendly, similar to how Apple has done with previous technologies. Ternus' ability to deliver such a product is uncertain, but it's seen as a necessary step for the company to stay ahead in the AI transformation.

Tim Cook's legacy as Apple's CEO is marked by innovations like AirPods and Touch Bars, but his recent step down has paved the way for John Ternus to take the reins. This transition may lead to significant changes at the tech giant. Cook's departure marks the end of an era at Apple.

Apple CEO Tim Cook is stepping down in September, to be succeeded by hardware chief John Ternus, marking a significant transition for the tech giant. Ternus will inherit a durable business but also face challenges such as pressure on the App Store's 30% cut and shifting developer dynamics. The change is expected to have implications for startups and the broader tech ecosystem.

DeepSeek has unveiled its new large language model, DeepSeek V4, which includes two versions, V4 Flash and V4 Pro, boasting improved efficiency and performance. The models have almost closed the gap with current leading models on reasoning benchmarks, but slightly lag behind in knowledge tests. DeepSeek V4 is more affordable than its frontier model counterparts, with the V4 Flash model costing $0.14 per million input tokens and $0.28 per million output tokens.

Intel is betting on AI inference to restore the CPU to the center of compute, with CEO Lip-Bu Tan stating that the total addressable chip market is pushing towards $1 trillion. The company hopes to capture a share of this market with its products, particularly in the areas of inference and agentic workloads. Intel reported Q1 revenue of $13.6 billion, with AI-driven business lines accounting for 60 percent of that figure.

Cybercriminals are utilizing AI to create sophisticated scams, including turbocharged phishing and hyperrealistic deepfakes, making it challenging for organizations to cope with the volume of cyberattacks. Meanwhile, AI is being used in healthcare to aid doctors with tasks such as notetaking and interpreting medical exam results, but its impact on patient outcomes remains unclear. The integration of AI in these fields is transforming the way they operate, but also raises important questions about effectiveness and security.

Meta has signed a multi-year collaboration with Amazon Web Services to deploy tens of millions of Graviton 5 CPU cores, supporting its AI ambitions and diversifying its compute fleet. The move is part of a larger trend towards adopting Arm-based CPUs for AI applications. Meta's partnership with AWS will help the company scale its infrastructure and reduce dependence on traditional x86 architectures.

Elon Musk is suing OpenAI, a company he co-founded, with a trial set to start on April 27th in Oakland, California. The lawsuit alleges OpenAI defrauded Musk, but the case is expected to reveal more about the messy power dynamics between Musk and current OpenAI CEO Sam Altman. The trial will likely uncover details about the inner workings of OpenAI and the feud between Musk and Altman.

Meta has signed a deal with Amazon to use millions of AWS Graviton chips to power its growing AI needs, marking a significant win for Amazon's homegrown chips. The deal brings more of Meta's cash back to AWS instead of competitors like Google Cloud. This partnership showcases Amazon's ability to provide AI-related compute needs with its custom-designed chips.

Open source models can find bugs as effectively as Anthropic's Mythos, according to OpenAI's first security hire, Ari Herbert-Voss, who suggests that building 'scaffolding' to run multiple open source models can achieve comparable results. Herbert-Voss argues that open source alternatives are not just viable but necessary for many organizations due to the high cost of Mythos. Human expertise is still needed to orchestrate open source models and assess AI-generated bug reports.

Chinese AI company DeepSeek has previewed its new AI model V4, which can compete with leading US rivals and marks a major improvement in coding capabilities. The model is also compatible with domestic Huawei technology, a milestone for China's chip industry. DeepSeek's release comes a year after it first gained attention for its AI capabilities.

The increasing use of AI in healthcare has raised questions about its effectiveness in improving patient outcomes, with many providers not rigorously assessing how well these technologies work. While AI tools can deliver accurate results, there is limited research on their impact on patient health. Researchers are calling for more studies to evaluate the effectiveness of AI in healthcare settings.

AI models are utilizing real-time cryptocurrency data to interpret market behavior, providing valuable insights into the rapidly changing financial landscape. The use of real-time data allows AI systems to spot changes and respond quickly, making them more effective in non-linear environments. This approach enables AI models to better understand complex market signals and make more informed decisions.

Claude, an AI model, is being updated to ensure impartiality and prevent political bias ahead of the US midterms and other global elections, with measures including character training and system prompts. The model is evaluated for its ability to provide balanced responses and is tested against attempts at misuse. Claude's developers are also working with external organizations to review model behaviors around freedom of expression.

Anthropic and NEC Corporation are collaborating to build Japan's largest AI engineering workforce, with NEC set to utilize Anthropic's Claude AI model to develop industry-specific AI products for the Japanese market. The partnership aims to create secure and reliable AI solutions for sectors like finance, manufacturing, and local government. Approximately 30,000 NEC Group employees worldwide will have access to Claude.

DeepSeek-V4 has been released, featuring a million-token context window and efficient large context length support, making it suitable for agentic tasks. The new architecture achieves this through hybrid attention mechanisms, including Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA). This innovation enables agents to handle long-running workloads without breaking, making it a significant development in the field.

The Mac mini has sold out on Apple's website due to high demand for running on-device AI models, leading to a surge of overpriced models on eBay. The shortages have extended to various configurations, with some models not shipping until June. As a result, prices for used and refurbished Mac minis have increased significantly.

Gemini, an AI assistant, can help with spring cleaning and home organization by generating custom checklists, troubleshooting repairs, and maximizing storage space. It can also assist with managing inboxes, planning errands, and providing plant care advice. With eight tips, Gemini simplifies tasks such as decluttering, cleaning, and home makeovers.

A 40-year-old man in South Korea faces up to 5 years in prison for using AI to create a fake image of a runaway wolf, obstructing an urgent investigation. The wolf, named Neukgu, had escaped from a zoo and its safe return was crucial for a wolf population revival effort. The man's actions caused concern and hindered the rescue efforts.

Uber CTO Praveen Neppalli Naga will join the StrictlyVC San Francisco event on April 30 to discuss operating at scale in the age of AI. The event, which features a stacked lineup of speakers, is a key opportunity for founders and investors to network and learn from industry leaders. Naga's conversation will cover his experience building complex systems at Uber and his focus on developing earnings systems for drivers and couriers.

The founder of Open Telemetry, Ted Young, aims to make the project 'boring' by stabilizing its elements to achieve graduation as a CNCF project. To do this, he plans to utilize AI tools to handle the scale of instrumenting software. The goal is to increase automation and reduce the load on maintainers.

Nothing has introduced Essential Voice, an AI-powered dictation tool that turns speech into formatted text and supports over 100 languages, available on select devices with more to follow. The feature allows for custom voice shortcuts and can translate text directly from one language to another. It is currently available on the Phone (3) with rollout planned for other devices.

Microsoft is offering voluntary buyouts to approximately 7% of its US staff, targeting experienced workers, in an effort to improve quality and reliability of Windows, despite potentially losing valuable skills. The move comes after the company has laid off thousands of employees in recent years. This decision may hinder Microsoft's goal to address quality issues in its software.

The World Press Photo of the Year award has been announced, with 'Separated by ICE' by Carol Guzy taking the top spot, sparking conversation about the role of AI in photography. The competition had strict rules around AI tool usage. The award celebrates the best of photojournalism, emphasizing the capture of reality.

A group of creators are making AI-generated influencers, primarily targeting a gay male audience, with characters like Jae Young Joon gaining thousands of followers on Instagram. Despite disclosing that the characters are AI-generated, many followers choose to engage with them as if they were real. The phenomenon has sparked discussions about the potential for deception and the impact on body standards in the gay community.

While chatbots like ChatGPT can be helpful for exploring financial options and understanding concepts, they should not be relied upon for professional financial advice due to limitations such as confidently outputting incorrect answers and affirming preexisting beliefs. Chatbots require sensitive information to provide tailored results, which can be a concern for users. It's essential to approach chatbots with skepticism when seeking money tips.

The latest advancements in coding agents are accelerating various types of software work, with frontend development seeing the most significant speedup, while backend development, infrastructure, and research are accelerated to lesser degrees. Coding agents are fluent in popular frontend languages and frameworks, but struggle with visual design and complex infrastructure decisions. As a result, expectations for software teams are shifting, with frontend teams expected to implement products faster than before.