David Sacks has concluded his tenure as Donald Trump’s AI and cryptocurrency adviser.

In a conversation with Bloomberg on Thursday, the seasoned entrepreneur, investor, and podcaster confirmed that his non-consecutive 130-day role as a special government employee has ended and that he’s transitioning to co-chair the President’s Council of Advisors on Science and Technology (PCAST), partnering with senior White House technology adviser Michael Kratsios. 

“I believe that as co-chair of PCAST, I can now provide insights not only on AI but on a broader array of technology issues,” he mentioned to Bloomberg during a video interview. “So yes, this is how I’ll be involved from now on.”

Practically, this signifies that Sacks will have less proximity to the power dynamics in Washington than he has had since the beginning of this second Trump term. As the AI czar, Sacks maintained direct communication with Trump and influenced policy decisions. PCAST serves as a federal advisory committee, meaning it investigates matters, produces reports, and forwards recommendations but does not create policy.

The council has been around in various forms since FDR, yet Sacks emphasized to Bloomberg that this particular group boasts “the most star power of any similar assembly” ever formed, and it’s challenging to dispute his assertion. The initial 15 members encompass Nvidia’s Jensen Huang, Meta’s Mark Zuckerberg, Oracle’s Larry Ellison, Google co-founder Sergey Brin, Marc Andreessen, AMD’s Lisa Su, and Michael Dell, among others. 

That’s quite a collection of billionaires.

Sacks informed Bloomberg that the council will address AI, advanced semiconductors, quantum computing, and nuclear energy, with immediate focus directed toward advancing Trump’s national AI framework, unveiled just last week. The framework aims to replace what Sacks referred to as a chaotic mix of conflicting regulations at the state level. “You’ve got 50 different states regulating this in 50 different ways,” he stated, “and it’s resulting in a patchwork of regulations that our innovators find challenging to navigate.” 

What Sacks did not directly address was the reasoning behind the timing of this transition and whether his recent remarks influenced it. Earlier this month, during the well-known “All In” podcast he co-hosts, Sacks publicly encouraged the administration to seek an exit from the U.S.-backed conflict with Iran, outlining a series of deteriorating scenarios — attacks on oil infrastructure in neighboring nations, the destruction of desalination facilities, the potential for nuclear engagement by Israel — and advocating for a diplomatic exit. Trump replied by informing reporters that Sacks had not discussed the war with him. 

When Bloomberg inquired about it on Thursday, Sacks figuratively raised his hands in disbelief: “I’m not part of the foreign policy or national security teams,” he remarked, specifying that his podcast commentary represented his personal perspective, not an official stance.

Despite the notable individuals Sacks is bringing to PCAST, it’s pertinent to consider the historical context of what the council has been, functioning as an advisory group with significant influence in some administrations and negligible impact in others. 

President Obama’s administration is remembered as the most productive on record, generating 36 reports over eight years — two of which resulted in tangible policy changes, including a regulation from the FDA that facilitated the availability of over-the-counter hearing aids. 

In contrast, President Trump’s initial council took nearly three years to appoint its first members, delivered a few reports, and did not leave a significant impact, while President Biden’s council leaned heavily towards academic expertise — featuring Nobel laureates, MacArthur fellows, and National Academy members — and produced a limited set of reports before the administration concluded. 

The current PCAST represents an entirely different entity, formed predominantly from the executive leadership of companies that are shaping the technology it will provide guidance on.

Now, Sacks is once again part of that landscape and likely free to move back into his role as an investor and entrepreneur. A representative for Craft Ventures, the firm Sacks co-founded and where he continues to be a partner, has not yet addressed related inquiries, but TechCrunch reported last year about the ethics waivers Sacks secured to retain financial interests in AI and crypto firms while influencing federal policy in both realms — an arrangement that received considerable criticism from ethics scholars and legislators.

A federal judge has favored Anthropic in its complex legal encounter with the Trump administration, granting the tech firm a legal injunction against the government’s latest order that classified it as a “supply chain risk,” according to the Wall Street Journal.

On Thursday, Judge Rita F. Lin of the Northern District of California instructed the Trump administration to revoke its recent categorization of Anthropic as a security threat, along with withdrawing its mandate that federal agencies sever connections with the company.

“It appears to be an effort to undermine Anthropic,” Lin reportedly remarked during the judicial proceedings. Lin ultimately contended that the government’s directions had violated free speech safeguards for the company.

The conflict between the Pentagon and Anthropic intensified last month over a disagreement about the government’s use of the AI company’s software. Anthropic had allegedly aimed to impose specific restrictions on how the government could utilize its AI models, including prohibiting their application in autonomous weapons systems or mass surveillance. The government opposed those restrictions, ultimately designating the company a supply chain risk—a label typically reserved for foreign entities. President Trump additionally instructed federal agencies to sever ties with the firm.

Shortly thereafter, Anthropic filed a lawsuit against the agency, along with Hegseth.

The White House has spent the past few weeks criticizing the company, describing it as “a radical-left, woke company” that threatens America’s “national security.” Anthropic CEO Dario Amodei, on the other hand, has branded the Defense Department’s actions as “retaliatory and punitive.”

Following Judge Lin’s ruling, Anthropic provided TechCrunch with the following statement: “We thank the court for acting promptly, and we are pleased they concur that Anthropic is likely to prevail on the merits. While this lawsuit was essential to defend Anthropic, our customers, and our partners, our emphasis continues to be on collaborating constructively with the government to ensure that all Americans benefit from safe and dependable AI.”

TechCrunch has also reached out to the White House for a response.

In the realm of AI chatbots, a fierce competition for user engagement is ongoing. Major chatbot companies are vying to boost their user numbers, and in a minor triumph, Google has recently simplified the process for users of rival chatbots to switch to Gemini.

On Thursday, the firm unveiled what it refers to as “switching tools,” new widgets crafted to enable users to migrate “memories” (essentially snippets of personal data) and even full chat histories from other chatbots straight into Gemini. According to the company, users can readily share “important preferences, connections, and personal context” using this method.

The intention is to significantly ease the transition to Google’s AI assistant, as users won’t need to invest extensive time retraining Gemini regarding their identity and desires.

The memory feature functions like this: Gemini will propose a prompt for the user to enter into their existing chatbot, which will subsequently produce a response that can be copied and pasted back into Gemini. In this manner, Gemini advises the user on what types of information would be beneficial for it to know about them, while also aiding in the transfer of that data back into its own system.

“Once you import these memories, Gemini will grasp the same essential details you’ve provided to other applications, such as your hobbies, your sibling’s name, or the town where you were raised,” the company states. “Rather than starting anew, you can swiftly bring Gemini up to date on what is most significant to you.”

For importing chat histories, Google claims that all it takes is uploading them in a zip file. It’s fairly straightforward to export chat logs in zip format from the majority of chatbots — including ChatGPT and Claude. This enables users to “seamlessly continue exactly where you paused,” according to the company. Google also mentions users can search through these past conversations.

ChatGPT continues to dominate the consumer chatbot landscape, with OpenAI sharing last month that it has reached 900 million active users each week. Gemini — despite Google’s extensive distribution advantages, including its default status on Android devices and the Chrome browser — has struggled with consumer recognition. Last month, it disclosed its own metrics during Alphabet’s fourth-quarter earnings report, revealing that Gemini had exceeded 750 million monthly active users. This strategy is clearly aimed at aiding Google in narrowing the gap.

As artificial intelligence penetrates the realms of editorial and media, platforms are rushing to set foundational guidelines for its utilization. This week, Wikipedia prohibited the incorporation of AI-generated text by its editors — yet it refrained from imposing a complete ban on AI within the site’s editorial framework.

In a recent adjustment to its guidelines, the platform now articulates that “the application of LLMs to create or modify article content is forbidden.” This updated wording refines and clarifies earlier, less clear language that indicated LLMs “ought not to be employed to produce new Wikipedia articles from the ground up.”

The involvement of AI in Wikipedia articles has sparked a heated debate within the extensive, volunteer-based community of editors. 404 Media reveals that the updated guideline, which was voted on by the editors of the site, received overwhelming backing — 40 to 2.

Nevertheless, the revised guidelines still allow for the ongoing use of AI in certain editorial tasks.

“Editors are allowed to utilize LLMs to propose basic copyedits to their writings, and to incorporate some suggestions post human evaluation, as long as the LLM does not inject any content of its own,” the revised policy specifies. “Caution is necessary, since LLMs may exceed your requests and alter the meaning of the text to a degree that is not substantiated by the cited sources.”

Mastodon is implementing modifications aimed at enhancing the attractiveness and usability of its social networking service, particularly for mainstream users seeking alternatives to X or Threads.

On Thursday, the creator of decentralized social networking software announced it is overhauling a crucial element of its platform by providing people’s user profiles with a fresh appearance, which it hopes will attract both organizations and individuals.

Founded on the ActivityPub protocol, Mastodon gained prominence after Elon Musk took over Twitter, now rebranded as X, prompting some users to look for alternatives. The platform’s allure lies in its decentralized nature, meaning no single entity controls the algorithm, and users can transfer their accounts should they disapprove of a specific server’s operations or community management.

Nevertheless, this arrangement is also more complex compared to joining a traditional social network like X. On Mastodon, users must select a server to participate in and experience different timelines (local and federated), which can be perplexing for newcomers. The procedure for following other users on the service can also be tricky.

This has left Mastodon facing challenges in attracting new users, with numbers currently around 800,000 monthly active users, down from a million during the peak of the Twitter saga.

In recent months, Mastodon has been addressing several challenges that might deter users. In February, it streamlined the onboarding experience and incorporated additional features that users typically expect, such as Quote Posts and “starter packs” known as Collections.

Now, it is focusing on user profiles. The revamped edition introduces several alterations, many of which are visual.

What’s changing

Rather than presenting two views of a user’s posts (“posts” or “posts and replies”), akin to X, profiles now contain a single “Activity” tab equipped with a dropdown menu. This enables users to customize different combinations of posts by enabling or disabling replies and boosts — with boosts being Mastodon’s equivalent of a repost.

Hashtags are now prominently displayed at the top of the Activity tab, enabling users to filter posts on that account by the selected tag.

Mastodon has also eliminated the pinned posts carousel, a feature that many users found unappealing. This was created to reconcile the requirements of those wanting to pin multiple posts with those visiting a profile to quickly access the user’s latest posts. Now, users with several pinned posts will showcase one prominently, while others can be accessed via a new “View all pinned posts” button.

Another alteration aims to clarify Mastodon handles for new users. In contrast to X or Threads, where users are simply @username, Mastodon handles incorporate two @’s — one indicating their account name and the other their server’s name. A new information pop-up elucidates this.

Users are also given greater control over their profile’s appearance, with options to conceal the “Media” or “Featured” tabs, if desired, or to hide replies from their “Media” tab to highlight their work.

Custom fields on the profiles where users can add elements like links, pronouns, and additional details are now arranged horizontally, creating more vertical space on the screen. These fields can now also be updated on iOS and Android, not solely on the web.

Additional design adjustments make profiles appear less cluttered — such as the elimination of a “following you” badge and repositioning the optional “personal note” users add to their profiles into an overflow menu.

Profile modifications can now all be managed from a single location in the account settings, enabling users to oversee tasks such as their featured hashtags (which Mastodon now thoughtfully suggests), links, and other profile details.

Link verification — a feature of Mastodon to ascertain a user’s trustworthiness without establishing a centralized authority (or demanding payment like X) — is now more accessible rather than being hidden in settings. Users can crop and add alt text to their profile images and cover photos.

These updates will be initially accessible to the mastodon.social server and other servers that choose to operate the nightly build. More servers will receive the update with the launch of the Mastodon 4.6 software version in a few weeks.

Netflix has discreetly increased its prices once more. The streaming leader’s lowest-priced, ad-supported plan is now $8.99 monthly, up from the earlier $7.99 subscription fee, Netflix confirmed to TechCrunch in a message.

The standard ad-free plan now costs $19.99 per month, reflecting a $2 rise from the former $17.99 subscription charge, while the premium option is also escalating by $2 and will now be priced at $26.99 per month.

It’s also becoming pricier to add additional viewers outside your residence. Adding a user to an ad-supported plan now incurs a fee of $6.99, instead of the previous $7.99. For ad-free plans, adding an extra viewer will now cost $9.99 rather than $8.99.

The company informed TechCrunch that these modifications aim to reflect enhancements to its “diverse selection of entertainment” and the quality of its services.

The price increases were initially noticed by Android Authority.

Netflix states that new subscribers enrolling will experience the adjusted plan costs from March 26, while current users will see the new prices implemented over the upcoming months. Current members will receive an email notification a month prior to the new rates taking effect.

Netflix last raised its prices in January 2025. Since that time, the platform has introduced a variety of new features, including the debut of video podcasts and an increase in livestreaming offerings. The company has also recently revealed plans to enhance its mobile application and expand its short-form video functionalities.

These new increases follow Netflix’s decision last month to withdraw from an acquisition attempt for Warner Bros. Discovery.

Warner Bros. Discovery announced that Paramount Skydance’s offer of $31 per share was a “better proposal” and had given Netflix four business days to respond. Netflix subsequently declared it would not increase its $82.7 billion all-cash proposal for the studio, ultimately abandoning the deal.

Once again, AI dominated the conversation at the latest YC Demo Day showcasing. Almost 190 startups were part of Y Combinator’s Winter ’26 cohort and presented their innovations during a Demo Day held on Tuesday. 

These startups are developing products in sectors like law, transportation, and healthcare.  

I must confess, I did not manage to listen to every single pitch due to the large size of the cohort and the format available for this year’s Demo Day to the media. YC released the pitch videos sequentially, approximately 20 minutes after the founders’ presentations (instead of a livestream or an invite-only session).  

Instead, I researched all 190 startups that presented and spent the day watching pitches from those that captured my interest, eventually narrowing it down to the 16 that emerged as the most noteworthy startups from this overloaded YC cohort. 

ARC Prize Foundation
What it does: Establishes benchmarks to assess progress toward AGI.  

Why it’s interesting: A nonprofit in YC! Yet, when organizations like OpenAI, Anthropic, and Google utilize some of its benchmarks, it becomes clear why it was included. This foundation is dedicated to fostering open-source AGI research by organizing competitions and granting research funds. A significant motivation behind this AI revolution is to achieve AGI (which Nvidia’s CEO Jensen Huang asserts has already been reached), and it will become a historical reference for monitoring how close we are to AI systems possessing general intelligence. 

Asimov
What it does: Gathers human movement data to train humanoids  

Why it’s interesting: People globally submit videos of themselves executing movements and tasks to this company, which converts them into datasets useful for training robots. This is part of a movement striving to make humanoids more functional, exploring applications beyond supply chains and entertainment. I am optimistic about humanoid technology, even if our “Rosey the Robot” age may still be far off. Leveraging data to instruct humanoids on the nuances and — dare I say, grace — of human movement could help them appear less robotic while performing tasks.  

Avoice
What it does: Streamlines the monotonous non-design tasks for architectural firms  

Why it’s interesting: It’s not routine to encounter new tech aimed at the architecture field. The founders have noted that this market is underrepresented (albeit rich in potential). This tool employs AI to automate tasks often viewed as tedious by creative professionals like architects, such as reviewing specifications, drawings, contracts, and proposals.  

Button Computer
What it does: A wearable AI  

Why it’s interesting: As the world anticipates OpenAI’s product from its acquisition of Johnny Ive’s company, everyone is banking on wearable AI. Two former Apple employees (may I refer to them as that?) have united to introduce Button, essentially a diminutive computer, as explained by the founders, designed for AI. Button connects with apps such as email, Slack, and Salesforce, operating them through voice commands to execute specific tasks. The next essential hardware will likely take the form of an AI wearable, making it fascinating to observe the emerging innovations.  

CodeWisp 
What it does: Enables anyone to create games using AI  

Why it’s interesting: The founders assert that you simply instruct an AI on creating a game, and it will do so. That’s fun, imaginative, and exhilarating! As a child, I attempted to design games but found the process challenging and tedious, yet the thrill of creating one never dimmed. While vibe coding has surged in popularity for app development, tools like this simplify the creative execution considerably. Perhaps this will usher in the next phase in vibe creation.  

Crosslayer Labs
What it does: Facilitates the detection of website spoofing  

Why it’s interesting: The emergence of agentic tools has made it simpler for malicious entities to spoof websites, with scammers readily exploiting this technology. Crosslayer Labs assists its clients in identifying and monitoring their online environments to safeguard against this shade of rising internet threats.  

Doomersion  
What it does: Teaches languages while you doomscroll 

Why it’s interesting: We spend excessive amounts of time doomscrolling, cramming our brains with content that often annoys or dulls our mental faculties. This startup introduces an app that presents users with short videos, akin to how they would scroll on TikTok, in the language they wish to acquire. If there’s to be any significance to the content we flood into our minds throughout the day, what a clever way to achieve that. It merges something people can’t resist (hours spent on their devices swiping through content) with something as appealing as language acquisition. Très intéressant. 

Lexius
What it does: Incorporates advanced AI into security systems  

Why it’s interesting: This leverages AI to enhance existing security camera systems, allowing for detecting and reporting theft or falls, replacing a disjointed, primarily manual process. The startup claims it focuses on businesses utilizing cameras lacking AI capabilities, where while a camera may capture an event, companies face delays in taking necessary actions. 

Librar Labs
What it does: An AI-driven library management system tool  

Why it’s interesting: This is AI entering an industry often overlooked by tech — libraries. This startup has developed an AI-driven library management system aimed primarily at schools for inventory and cataloging purposes. As the founder emphasized during his pitch, competition in automating or innovating existing tools in this sector is minimal, making any new concept a candidate for “the next big thing.”  

Milliray
What it does: Radar system designed to track small drones  

Why it’s interesting: Defense technology is currently one of the most sought-after fields in tech. The founder of this venture stated that, currently, individuals are on-site doing everything possible to trace tiny drones, but human eyesight can overlook or confuse a small drone with a bird and vice versa. This startup employs sensors to accurately identify a tiny drone in the air. In light of current geopolitical circumstances, new technologies are essential — and unfolding — to ensure every nation stays proactive against potential threats.  

MouseCat
What it does: Utilizes AI to investigate fraud  

Why it’s interesting: AI serves as a valuable asset in professional tasks, and that rings true for individuals involved in fraud and scamming. This firm extracts data from large cloud sources like Databricks or Snowflake, evaluates consumer information and activities for any red flags, and provides recommendations on appropriate actions to take. AI-native tools such as this are crucial for keeping pace with the adverse implications that malicious AI can also incite.

Opalite Health
What it does: Employs AI to assist healthcare providers in communicating with non-English speakers 

Why it’s interesting: When two individuals can’t comprehend each other, much is left to interpretation. In the realm of healthcare, that could result in life-threatening situations. This AI medical translator facilitates the breakdown of language barriers, allowing healthcare providers to understand patients speaking different languages. In a global society (and in a diverse country like America), it’s crucial for individuals to receive necessary healthcare, regardless of language. Naturally, this concept isn’t entirely original, as several other startups and health tech firms provide similar solutions. 

Sequence Markets
What it does: Enables trading across diverse markets, such as crypto and predictions, within a single system.  

Why it’s interesting: As someone who prefers having everything consolidated, I appreciate the desire for a less fragmented trading experience in these markets. It’s akin to why I enjoy going to large retailers to view various brands (I like having all my options visible at once).  

ShoFo
What it does: A comprehensive video library 

Why it’s interesting: This startup markets itself as the “world’s video library,” which I find quite fascinating. Having grown up with YouTube and Tumblr, I recall how challenging search functionality was back then when looking for something specific. Although this serves as more of a custom video index to aid AI Labs in efficiently sourcing diverse datasets, I appreciate any tool that simplifies search and organization.  

Sonarly
What it does: Assists software in rectifying its production challenges  

Why it’s interesting: This startup is pioneering technology that sounds particularly impressive. It integrates with other monitoring systems, promising to minimize alert noise (a distraction from essential alerts), autonomously identifies root causes of issues, and proposes solutions or next steps for engineers. While there’s a growing emergence of AI code review startups (and that feature is becoming available from model creators), there should still be an opportunity for independent solutions once the code is deployed in production environments. This is yet another workflow aspect that founders are streamlining.  

Terranox AI
What it does: Employs AI to discover uranium deposits in North America   

Why it’s interesting: Uranium will be essential for driving the next generation of nuclear energy, as the founders of this venture pointed out. (Nuclear energy is generally regarded as safe today, though uranium itself is, of course, harmful — and the founder did not mention how it could be excavated safely.) These founders believe that nuclear energy will play a crucial role in supplying power to the numerous new data centers being established. The planet will need a significant amount of energy, from various sources, to support human aspirations amidst the AI revolution.  

On Thursday, Google revealed that its AI-driven “Live Translate” feature within Google Translate, which allows users to listen to real-time translations through their headphones, is now extending its availability to iOS and additional countries.

With this rollout, the feature can now be accessed on both iOS and Android in the U.S., India, Mexico, Germany, Spain, France, Nigeria, Italy, the United Kingdom, Japan, Bangladesh, and Thailand. Previously, it was exclusively on Android for users in the U.S., India, and Mexico.

This functionality effectively transforms any headphones into a one-way translation tool that operates in real time. According to Google, the real-time translation experience, powered by Google’s Gemini AI, preserves the tone, emphasis, and rhythm of each speaker. This enhances understanding of the dialogue and clarity regarding who is speaking.

The tech giant encourages users to leverage this feature for various scenarios, such as engaging in dinner conversations with relatives who speak a different language, or comprehending train announcements while in a foreign country.

This feature is compatible with any headphones and offers support for over 70 languages.

Users can initiate the feature by launching the Google Translate app, selecting the “Live Translate” option, and then linking their headphones.

The extension of Live Translate coincides with Google’s announcement of the global expansion of its AI-driven conversational search feature, Search Live, to all languages and regions where AI Mode is accessible. This means individuals in more than 200 countries and territories will now have access to the feature that was previously limited to the U.S. and India.

Launched in July 2025, Search Live enables users to direct their phone camera at objects for immediate assistance, facilitating dynamic conversations informed by visual context from the camera’s view. To access this feature, users should open the Google app on either Android or iOS and tap the Live icon found beneath the Search bar.

OpenAI has halted yet another initiative — at least temporarily. On Thursday, the Financial Times disclosed that the AI organization would be pausing plans to create an “erotic” version of ChatGPT “indefinitely.”

The suggested “adult mode,” initially proposed by CEO Sam Altman in October, stirred significant controversy among technology oversight groups and even within OpenAI’s own team. In January, a discussion between company leaders and their advisory council became intense, with one adviser warning that OpenAI could be on the brink of developing a “sexy suicide coach,” as reported earlier by The Wall Street Journal.

In light of all the backlash, the launch of the feature faced several delays.

FT mentions that there is currently no schedule for the erotic feature’s release. When approached for comments by TechCrunch, an OpenAI representative stated that the company had “nothing more to add.”

Adult mode is merely the latest endeavor that OpenAI has shelved in recent days as the AI behemoth sharpens its focus. On Tuesday, the firm discreetly revealed it would be deprioritizing Instant Checkout, a feature in ChatGPT aimed at transforming the chatbot into a purchasing platform for items from e-commerce sites. Following that, on Wednesday, the company unexpectedly announced the closure of Sora, its AI video generator, which had faced criticism for contributing to the surge of low-quality AI content online since its debut in 2024.

These adjustments come about a week after The Wall Street Journal reported OpenAI would embark on a “significant strategy shift” to move away from diversions in order to concentrate on its core objectives: business clientele and programmers.

Why has OpenAI chosen to eliminate distractions at this specific time? It may be responding to pressure from Anthropic, which has been diligently launching a range of coding and business tools in recent months — and has achieved notable success in attracting customers as a consequence.

The two firms have also been openly clashing over Pentagon contracts — a contest that seems to have favored OpenAI. Three weeks prior, it announced a $200 million deal with the Department of Defense, while Anthropic is currently embroiled in a legal struggle with the agency.

In summary, if recent events indicate anything, the trajectory of AI appears to be shifting away from adult content and memes toward a focus on business and defense.