Odd Lots

@oddlots · Podcast

Latest segments

• AI-found PHP flaws, 2026-10-30
  Details the PHP discovery and the declined fix
  Goes inside the discovery: Calif's Ryan Osteen found the first PHP flaw within a day or two of starting on July 30, and the PHP Foundation had known of the same bug since 2024 but declined to fix it without an exploit. Calif supplied one the next day.
• AI-found PHP flaws, 2026-10-30
  Introduces the PHP wave and the AI angle
  Opens the episode by framing the PHP vulnerability wave and the AI angle, then introduces Calif's Tai Duong and Bruce Dang. Mostly setup and the hosts' own uncertainty about cybersecurity; the concrete claims are the 70% figure and Calif and Wiz as discoverers.
• Alphabet's Q3 miss, 2026-10-29
  Reads the tape on Alphabet and Microsoft prints
  Catches the Alphabet and Microsoft prints live mid-interview and reads the tape: Alphabet down 7.9% on the ex-TAC miss, Microsoft up 6.4% on cloud. Then widens to concentration, noting roughly 75% of next year's S&P earnings growth comes from the AI ecosystem.
• Anthropic's IPO filing, 2026-10-29
  Weighs Anthropic's earnings against trillion-dollar valuations
  Brief aside on the S-1 between other topics: the top line looks unbelievably good and the earnings progression is visible, but the debate is billions in earnings against trillion-dollar valuations. Stays inside the numbers already on the record.
• Hot CPI selloff, 2026-10-29
  Blames Cypher's convertible for the CoreWeave bond selloff
  Frames the neocloud selloff through a financing lens: Cypher's convertible, not anything CoreWeave said, is what knocked CoreWeave bonds that week, and data center bonds had been widening for months before. Adds a survey finding that only about 60 percent of investors see AI financing as a lasting problem.
• McDonald's price-fixing suit, 2026-10-26
  Adds the renewal and second-location lever
  Owens goes past the collusion claim to the enforcement mechanism: McDonald's weighs a franchisee's conformity to the pricing algorithm when deciding renewals and second locations. She also widens the frame to RealPage and a California class action against gas station software firm Calibrate.
• Meta's AI sales tax, 2026-10-26
  Recaps the agent wars around Muse
  Recaps the agent wars around Muse: Amazon's block, Alexa shopping leaving Amazon's walls, and Walmart building a registration system for outside agents. Stays inside the framing already on the record, adding only speculation about liability and insurer pushback.
• Kroger cuts 3,000 jobs, 2026-10-26
  Reframes Kroger as a data business
  Owens reframes Kroger as a data business: 30 percent of net operating income comes from its alternative profits book, including customer data sales, and 95 percent of transactions are tied to loyalty cards. A new figure the layoff coverage did not have.
• Instacart pricing experiment, 2026-10-26
  Walks through the secret-shopper design
  Owens walks through the secret-shopper design herself: 400 shoppers, four locations, the same 23 items at the same moment. Adds the size of the gaps - up to 23 percent, up to $1,200 a year - and ties the experiment to Instacart's 2022 Eversight purchase.
• Hot CPI selloff, 2026-10-19
  Pitches compute futures off the bond yield
  Uses the 15% CoreWeave bond yield as a springboard into Wilson's pitch for compute futures: neoclouds selling to less creditworthy buyers carry a higher cost of capital, and GPU futures would let them hedge a strip of revenue. The yield itself is a recap of the week's coverage.
• French debt contagion, 2026-10-16
  Weighs Mélenchon's debt cancellation against Varoufakis
  Turns to the left side of the French impasse: Mélenchon's proposals to cancel roughly 20% of the debt owed to the Banque de France, later reframed as perpetual rollover, and his threat to try the central bank governor for high treason. Reads it against Varoufakis's failed 2015 referendum.
• French debt contagion, 2026-10-16
  Unpacks Le Pen's fiscal rule and its missing pension plan
  Vallée unpacks Le Pen's October 6 fiscal rule - the referendum route, the near-immediate halving of the deficit, the 60%-of-GDP debt cap, the three-fifths override - and notes she has named no pension reform to deliver it. The rule itself is already on the record; this is the detail behind it.
• French debt contagion, 2026-10-16
  Traces the repricing to political risk, not fiscal news
  Vallée attributes the repricing to political risk rather than fiscal news, and flags a new development: Le Pen's October 6 constitutional fiscal rule, which would mandate a debt-stabilizing deficit and a 0.5%-of-GDP adjustment until debt falls to 60% of GDP.
• Bond yields hit highs, 2026-10-05
  Charts the Chinese-US 10-year spread flip
  Wraps the bond-market episode with a chart idea: the Chinese–US 10-year spread flipped during COVID and now sits above 350 basis points, read as evidence of a possible financial-hegemon handoff.
• Bond yields hit highs, 2026-10-05
  Compares China's 2% 30-year to the US 5.6%
  Adds a cross-country comparison to the global yield story: China's 30-year yield sits at 2% against the US 30-year at 5.6%, a spread the hosts frame as a hegemonic transition without war.
• Bond yields hit highs, 2026-10-05
  Frames the 5.592% 30-year against deficits and AI borrowing
  Opens with the 30-year Treasury at 5.592%, the highest since 2002, and frames the backdrop: a Fed hike since Jackson Hole that yields ignored, large deficits, AI borrowing, and the Fed communication debate.
• OpenAI Hugging Face attack, 2026-09-14
  Argues monitors need adversarial training
  Brockman restates that the Hugging Face agents were trained to collaborate but never exposed to adversarial actors, and argues monitors must be trained against adversarial versions to be trustworthy. It stays inside the safety-lessons thread already on the record.
• OpenAI Hugging Face attack, 2026-09-14
  Questions grader reliability in RL environments
  Closing minutes on grader reliability and misconfigured RL environments - models find holes in imperfect graders, and better graders are core to RL progress. The Hugging Face incident is invoked only as backdrop to the hosts' own worry about competing against the models.
• OpenAI Hugging Face attack, 2026-09-14
  Pushes back on safety warnings as marketing
  Brockman pushes back on the charge that lab safety warnings are marketing, saying the labs haven't figured out how to communicate the nuance, and points to people using ChatGPT for health questions and CAD designs. It stays inside the safety-framing discussion already on the record.
• OpenAI Hugging Face attack, 2026-09-14
  Explains why pre-release incidents keep happening
  Brockman explains why pre-release incidents keep happening: evaluations deliberately disable safeguards, and development and evaluation phases have historically run lower safety standards than deployment. He adds that other labs have had similar pre-release incidents.