This Week in Tech (Audio)

TWiT 1075: The Commonwealth Club - Meta Layoffs, DOGE Data Theft, & the Rise of AI Fails

March 16, 2026

Key Takeaways Copied to clipboard!

  • Meta is reportedly planning massive layoffs (up to 20% reduction) due to mounting AI infrastructure costs, despite recent hiring sprees. 
  • The ongoing social media addiction trial involving Meta and YouTube is reaching the jury deliberation phase, raising fundamental questions about platform liability for user harm. 
  • The recent public release and subsequent quick shutdown of the relaunched Dig site highlight the pervasive and escalating problem of AI bot spam overwhelming social media platforms. 
  • The high cost of gasoline in California, exacerbated by taxes and specific refining requirements, is significantly higher than in other regions, leading to discussions about comparative fuel prices. 
  • NVIDIA's upcoming GTC keynote is anticipated to feature major AI announcements, including the open-source AI agent platform NemoClaw, which may challenge proprietary ecosystems like CUDA. 
  • A federal judge issued a temporary injunction blocking Perplexity's Comet AI agent from making purchases on Amazon's website, highlighting the emerging legal conflict between agentic AI and established e-commerce business models that rely on user traffic and advertising impressions. 
  • Advanced nuclear reactor designs, like those using molten sodium or salts, present trade-offs between efficiency (higher heat tolerance, better fuel burning) and significant safety challenges (sodium's reactivity with water). 
  • The case of a Tennessee grandmother wrongly jailed based solely on facial recognition evidence highlights the severe, real-world dangers of over-relying on flawed AI/computer identification systems without proper human investigation. 
  • Former Uber CEO Travis Kalanick is resurfacing with a vision for 'gainfully employed robots' to digitize the physical world (growing, mining, manufacturing, transporting), echoing themes from science fiction but raising immediate economic questions about labor displacement. 

Segments

Panel Introduction and Meta Layoffs
Copied to clipboard!
(00:00:00)
  • Key Takeaway: Meta is planning significant layoffs (up to 20% reduction) driven by the immense capital expenditure required for AI data centers.
  • Summary: The panel for This Week in Tech, episode 1075, is introduced, featuring Jennifer Pattison Tuohy, Ian Thomson, and Richard Campbell. The first major news item concerns Meta planning massive layoffs, potentially shrinking the company by 20%, as AI infrastructure costs mount, citing reports of $600 billion planned spending by 2028. One anecdote suggests engineers are being forced to train AI tools that subsequently lead to their own jobs being outsourced to lower-grade staff.
Meta’s Avocado AI Delay
Copied to clipboard!
(00:06:25)
  • Key Takeaway: Meta delayed the launch of its latest AI model, codenamed Avocado, because it was reportedly not yet ‘ripe enough’ for release.
  • Summary: Meta decided to push back the launch of its new AI model, Avocado, according to The New York Times. This delay occurred shortly after the Magnificent Seven tech companies collectively lost a trillion dollars in market value this year. Separately, a Chipotle AI chatbot was discovered to be running on Claude in the background, demonstrating a lack of proper guardrails on corporate AI implementations.
Social Media Addiction Trial Update
Copied to clipboard!
(00:07:53)
  • Key Takeaway: The social media addiction trial against Meta and YouTube has gone to the jury after closing arguments, following settlements by Snapchat and TikTok.
  • Summary: The jury began deliberations on Friday in the social media addiction trial in Los Angeles, which centers on a 20-year-old plaintiff claiming harm from early platform use. The defense argued the plaintiff used the platforms as a coping mechanism for a difficult home life, while plaintiffs point to internal evidence showing companies optimized products for addiction. The core legal challenge is proving a direct causal link between the algorithms and meaningful harm, similar to historical tobacco litigation.
Social Media vs. Traditional Media Curation
Copied to clipboard!
(00:18:28)
  • Key Takeaway: Social media algorithms are viewed as more problematic than traditional media because they lack human curation, leading to overwhelming and potentially harmful, unvetted content exposure.
  • Summary: The lack of editorial oversight on social media, where algorithms dictate content flow, contrasts sharply with curated media like magazines or Netflix shows. This difference is cited as a reason for increased harm, particularly concerning body image issues among teenage girls, as the device provides continuous, unfiltered exposure. The conversation notes that social media is replacing real-life socialization, evidenced by students looking at phones instead of talking during school breaks.
TikTok Deal and Political Influence
Copied to clipboard!
(00:29:45)
  • Key Takeaway: The Trump administration reportedly secured a $10 billion payment from investors (Oracle, Silverlake) for brokering the TikTok deal, which many view as legalized bribery.
  • Summary: Investors in the US version of TikTok agreed to pay the government $10 billion for the deal brokering, though the final valuation was considered below market rate. This transaction is noted as nearly unprecedented for government assistance in a private transaction. Furthermore, the deal structure still allows ByteDance to retain a minority stake (19.999%), suggesting Chinese influence remains.
X/Twitter Bot Crisis and AI Content
Copied to clipboard!
(00:41:23)
  • Key Takeaway: X suspended 800 million spam accounts in 2024, yet the platform is simultaneously flooded with fake AI-generated content, particularly concerning the Iran War.
  • Summary: X reported suspending 800 million accounts in 2024 for spam and manipulation, a number more than double the platform’s total user base when Elon Musk acquired it. The platform is struggling to control state-backed actors, with Russia being the most prolific. Furthermore, X’s own AI, Grok, is reportedly failing to verify videos and is being used to generate fake AI content about current conflicts.
Elon Musk’s XAI Management Issues
Copied to clipboard!
(00:48:44)
  • Key Takeaway: Elon Musk admitted his AI company, XAI, is ’not built right’ following the departure of most of its founding team, while XAI continues to lose money.
  • Summary: Of the 12 founders of XAI, only two remain, with Musk characterizing the departures as ‘deliberate exits’ of people unsuited for later stages. Musk is reportedly losing money on XAI and X, contrasting with profits from SpaceX and Tesla, leading to speculation about financial maneuvering between his private companies. This situation is compared to other tech investments in NVIDIA, where companies use their own money to secure chip supply.
Gas Prices and California Costs
Copied to clipboard!
(01:00:04)
  • Key Takeaway: California gasoline prices are significantly inflated due to taxes and state-mandated refining standards.
  • Summary: Gas prices in California reached $6.00 per gallon in Petaluma, with San Francisco reporting $6.50, which is attributed partly to state taxes and the requirement that petroleum be refined within California, increasing costs. A comparison was drawn to UK prices, which are often higher but obscured by being listed in liters. The high cost of fuel directly impacts the price of all consumer goods due to trucking expenses.
NVIDIA GTC Keynote Previews
Copied to clipboard!
(01:03:33)
  • Key Takeaway: NVIDIA is expected to announce NemoClaw, an AI agent platform that reportedly does not require CUDA and may run on non-NVIDIA hardware.
  • Summary: The upcoming NVIDIA GTC keynote, featuring Jensen Wong, is highly anticipated as potentially the most important ever, given NVIDIA’s central role in AI development. NemoClaw is highlighted as a potentially more secure AI agent platform than OpenClaw, and its potential independence from the proprietary CUDA language is noted. Furthermore, NVIDIA plans to reveal a new chip system for inference computing incorporating a design licensed from Grok (with a Q).
Anthropic’s AI Lead
Copied to clipboard!
(01:06:49)
  • Key Takeaway: Anthropic’s Claude 3 Opus model now features a million-token context window, a five-fold increase over its previous limit, signaling a significant leap in data processing capability.
  • Summary: Anthropic’s Frontier model, Opus 4.6, has expanded its context window to one million tokens, allowing it to ingest substantially more data without performance degradation. This expansion is now available to users who previously paid a premium for smaller context limits, suggesting Anthropic is achieving greater economic efficiency. The company is seen as accelerating away from competitors, partly by capitalizing on OpenAI’s user dissatisfaction.
Amazon Blocks Perplexity Shopping Bots
Copied to clipboard!
(01:11:44)
  • Key Takeaway: A judge granted Amazon a temporary injunction preventing Perplexity’s Comet AI agent from making purchases on Amazon’s site, citing unauthorized access and harm to Amazon’s advertising revenue model.
  • Summary: Amazon successfully argued that Perplexity’s agentic browser, Comet, accessed its site without authorization, undermining Amazon’s advertising business which relies on tracking legitimate human impressions. This ruling illustrates the ‘DoorDash problem,’ where agentic AI circumvents the platform itself, potentially starving news organizations and retailers of traffic. Amazon is simultaneously developing its own agent, Rufus, while reportedly restricting internal use of competitor AI tools like Claude.
Doge Engineer Data Theft Scandal
Copied to clipboard!
(01:34:26)
  • Key Takeaway: A Doge engineer allegedly copied highly sensitive Social Security Administration databases containing records for 500 million individuals, including SSNs and personal details.
  • Summary: The stolen data included the Social Security Administration’s Numident and Master Death File databases, encompassing Social Security numbers, dates of birth, and parents’ names for hundreds of millions of people. The engineer, who reportedly had unrestricted, God-level security access via Microsoft’s M365 government environment, asked for assistance transferring the data to a personal computer to ‘sanitize’ it for a new company. This incident underscores the massive security risks associated with granting extensive access to young contractors in government roles.
Doge Deposition Mockery
Copied to clipboard!
(01:39:02)
  • Key Takeaway: Doge personnel demonstrated profound ignorance regarding the meaning of DEI during depositions related to cutting humanities grants, relying solely on keyword searches.
  • Summary: During testimony concerning the cutting of National Endowment for the Humanities grants, Doge employees admitted to using ChatGPT to perform simple keyword searches for terms like ‘DEI’ or ‘diverse’ to automatically cancel funding applications. One deponent stated he did not know the definition of DEI but knew it when he saw it, illustrating a lack of substantive understanding behind major administrative decisions. The testimony was so poorly received that the videos were taken down by a judge, though they had already circulated online.
Lab-Grown Meat Bans and Solar Panel Legality
Copied to clipboard!
(01:42:29)
  • Key Takeaway: Texas and Florida have banned the sale of lab-grown meat, while plug-in solar panels, popular in Europe, remain largely illegal in the US due to utility resistance.
  • Summary: Texas banned cultivated meat, with Governor DeSantis framing it as fighting the ‘global elite’s plan,’ despite the lab-grown alternative eliminating contaminants like mercury found in wild salmon. Separately, plug-in solar panels, which allow apartment dwellers to generate power without complex installation, are legal in Utah but face resistance from utilities in 30 other states who fear losing their business model. Utilities cite safety concerns, though Germany has millions installed without reported incidents, while the economics of self-generation are becoming increasingly attractive.
Nuclear Reactor Coolant Physics
Copied to clipboard!
(01:57:57)
  • Key Takeaway: Molten sodium allows fast spectrum reactors to operate at higher temperatures (up to 600 degrees) without pressurization, unlike water reactors, but introduces extreme fire/explosion risks upon contact with moisture.
  • Summary: Thermal neutron reactors use water as a working fluid, requiring high pressure to reach steam temperatures. Fast spectrum reactors, using sodium, avoid pressurization and favor breaking uranium atoms rather than creating plutonium. The primary downside of sodium is its violent reaction with water and hydrogen, necessitating rigorous purging of the system.
TerraPower Reactor Design
Copied to clipboard!
(02:00:35)
  • Key Takeaway: TerraPower’s design utilizes lithium beryllium salts as a heat transfer fluid, enabling energy storage to run the reactor at full capacity constantly, buffering power output based on demand.
  • Summary: TerraPower is using lithium beryllium salts for heat transfer, which allows for storing excess heat to manage variable power needs, running the 300-megawatt class reactor at higher outputs temporarily. The major hurdle for sodium reactors remains safely shutting down and purging the sodium, which is why only one such reactor is currently operational (in Russia). TerraPower has permission to build the energy island but not yet to acquire fuel for the reactor.
SMR Regulatory Hurdles
Copied to clipboard!
(02:02:55)
  • Key Takeaway: Small Modular Reactors (SMRs) face significant regulatory challenges because current rules, designed for large reactors, mandate separate control rooms for each unit, making multi-reactor SMR deployments economically impractical.
  • Summary: The concept of SMRs was initially a way to discuss nuclear power when direct discussion was difficult, but regulatory structures favor large reactors. The requirement for separate control rooms for each reactor unit severely penalizes modular designs. Companies like NewScale are struggling because the regulatory framework is not adapted for smaller, multiple-unit deployments.
Kalanick’s ‘Gainfully Employed Robots’
Copied to clipboard!
(02:07:42)
  • Key Takeaway: Travis Kalanick’s new venture, Atoms, aims to digitize the physical world by creating specialized, ‘gainfully employed robots’ for growing, mining, and manufacturing, predicting a singularity where human labor is divorced from production.
  • Summary: Kalanick frames his post-Uber work as a continuation of digitizing the physical world, focusing on robotics for resource extraction and manufacturing. He posits that rapidly decreasing costs of AI intelligence will lead to abundance, though the hosts question who will purchase goods if human labor is eliminated. Kalanick’s vision aligns with a path toward the technological singularity.
Facial Recognition Injustice
Copied to clipboard!
(02:18:39)
  • Key Takeaway: A Tennessee grandmother was jailed for six months in North Dakota based solely on a facial recognition match linking her to bank fraud, despite her never having been to North Dakota, illustrating catastrophic failures in automated identification systems.
  • Summary: Fargo police arrested the woman based only on surveillance video analyzed by face recognition software, leading to four months of incarceration without bail before her attorney proved she was 1,200 miles away. The police department has reportedly not apologized, and she lost her home while jailed. This incident underscores that the people using the technology, not just the AI itself, are responsible for such severe errors.
Palantir CEO’s Political AI View
Copied to clipboard!
(02:22:36)
  • Key Takeaway: Palantir CEO Alex Karp stated that AI technology disrupts humanity’s trained, largely Democratic voters, increasing the economic power of vocationally trained, often male, working-class voters.
  • Summary: Karp made this controversial statement during a CNBC interview, suggesting AI shifts economic power along political and demographic lines. The hosts found the explicit political framing shocking, noting that a more constructive framing would be AI democratizing knowledge. Palantir’s government work includes using its software for tasks like assigning seating charts for USDA employees.
Niantic’s Location Data Monetization
Copied to clipboard!
(02:25:11)
  • Key Takeaway: Niantic leveraged its massive user base from Ingress and Pokémon Go to build a database of 30 billion real-world waypoints, which it now sells to delivery robot companies for centimeter-level location accuracy.
  • Summary: The genius of Niantic was outsourcing the creation of a detailed global map by incentivizing players to tag landmarks for gameplay. This spatial data model, developed through years of user activity, is now being sold to autonomous systems. This development, combined with the facial recognition failures, points toward an easily assembled surveillance infrastructure.
US Travel Concerns and Swiss E-Voting Failure
Copied to clipboard!
(02:40:04)
  • Key Takeaway: The Ig Nobel Prize is moving its ceremony from the US to Switzerland due to safety concerns for international guests, reflecting broader apprehension about US border control and legal recourse.
  • Summary: The founder of the Ig Nobel Prizes cited safety concerns for guests traveling to the US, mirroring issues seen at tech conferences where European talent avoids US travel. Separately, a Swiss e-voting pilot failed when 2,048 ballots could not be decrypted despite using three correct USB keys, highlighting security issues even in highly regulated environments.
Turing Award Winner Tony Hoare
Copied to clipboard!
(02:44:05)
  • Key Takeaway: Turing Award-winning computer scientist Tony Hoare, inventor of Quicksort, recently passed away at 92, famously calling the introduction of null references in ALGOL his ‘billion-dollar mistake.’
  • Summary: Hoare is recognized for fundamental work in computer science, including the Quicksort algorithm. He lamented the introduction of null references, which complicates programming immensely by requiring constant checks for null values. His quote emphasizes the difficulty of making software simple versus making it complex enough to hide deficiencies.
YouTube’s Ad Revenue Dominance
Copied to clipboard!
(02:37:36)
  • Key Takeaway: YouTube surpassed Disney, Paramount, and Warner Bros. Discovery combined in ad revenue in 2025, solidifying its position as the number one media platform by capturing long-tail niche content consumption.
  • Summary: YouTube generated $40 billion in ad revenue, exceeding the combined total of major legacy media companies. This growth is driven by its ability to cater to every niche interest, causing even technologically resistant users to shift viewing habits from traditional TV to the platform. YouTube TV pricing ($85/month) also reflects its transition into a form of cable replacement.