Friend Acquisition Rate

Kalshi’s election markets gave Matt Levine a cleaner, more relaxing way to follow the midterms than the flood of punditry. He watched the odds of Democrats taking the Senate move from about 73% around 5 p.m. to 95% by 7:30 and 98% by 8:30, then went to bed. He described Kalshi as combining social-media signals and assessments of credibility in a Bayesian model that produces one probability. He considered the election-night service socially useful while retaining his criticism of the platform’s sports gambling.

A forecast can be well calibrated without being decisive for an individual decision: 73% still leaves substantial room for the other outcome, and even an 80% election-day probability implies a one-in-five chance of being wrong. One speaker cited Bolsonaro’s reported 80% odds before he lost Brazil’s election. The speakers also questioned claims that prediction markets let businesses hedge election risk. A governor’s election might proxy for a proposed data-center moratorium that could put tens of billions of dollars at risk, but the market might not have enough capacity to hedge an exposure that large. Forecasts could also affect voting, though the speakers said those effects are poorly understood and markets may be easier to manipulate than polls, including with fabricated polls intended to move prices.

One speaker said reading Kevin Roose’s AGI Chronicles and talking with people in AI had increased their concern about existential risk. They characterized Anthropic’s stated preference as slowing AI to reduce that risk, but its revealed preference as building data centers and advancing models quickly. They also argued that AI progress depends on fragile semiconductor expertise and manufacturing processes: disruption at a TSMC fab could set progress back substantially. AI might help design chips, but that would not necessarily replace workers’ tacit knowledge of how to adjust manufacturing machinery.

The collapse of Weil was presented as a run on a law-firm partnership and an example of how even a major firm can be fragile under mild financial distress. A paper by John Morley argues that stronger firms can help trigger such a collapse by poaching a weaker firm’s most lucrative partners. Weil ranked among the top 30 firms by revenue, and one speaker estimated average partner earnings the prior year at “five or $10 million.” Partners in private-equity M&A and fund formation can bring large teams of associates, clients, and work when they move; their departures can prompt others to leave. Law firms generally cannot use non-competes to prevent lawyers from practicing elsewhere and must be reasonable about returning departing partners’ equity, unlike banks, which can use non-competes, gardening leave, and deferred vesting. The speakers linked the vulnerability to firms’ growing reliance on lucrative private-equity work and to legal-ethics rules built around older assumptions of scarce lawyers and genteel, durable partnerships. As one FT journalist put it, law firms wanted hedge-fund pay but also got hedge-fund volatility.

AI may let rainmaking partners at financial-services firms do more work with fewer associates, potentially shrinking associate training. Another possibility is that clients use AI to answer legal questions directly and stop calling firms. One speaker said Wachtell’s possible unsettlement was probably not an AI story, though it might reflect broader anxiety. Its compensation was described as relatively lockstep, based more on seniority and firm-generated business than individual rainmaking; average partner profit was roughly $10 million, not $20 million. Wachtell reportedly earned about $90 million for closing Elon Musk’s Twitter deal, adding roughly $1 million to each partner’s profit. The speaker said they left despite liking its work, people, and culture because their health was breaking down; younger partners sometimes left for jobs paying $20 million elsewhere. Changing jobs can also help people make friends, though they may not keep old ones. Bloomberg’s high friend-acquisition rate was attributed to financial-media people coming there to record.

The Binance attackers were alleged to be North Korean, which North Korea denied. They reportedly spent months posing as interviewers before sending a malicious coding assignment. Roughly half the stolen assets were centralized stablecoins such as Tether and USDC, whose issuers could immobilize or cancel them. Tether and Circle reportedly acted about 40 minutes after the theft; before then, the attackers used decentralized exchanges to convert stablecoins into crypto, pushing Ether up about 15% with roughly 15% slippage. Trading firms said they made substantial profits taking the other side, but it was unclear whether they knew the funds were stolen. The speakers disagreed over whether reversing stablecoin transfers violated crypto’s permissionless, irreversible ethos or sensibly resembled banks reversing stolen transactions. They noted that recovery depended on senior contacts at Binance, Tether, and Circle coordinating by phone, and that much DeFi trading depends on centralized stablecoins. They described crypto use as largely trading financial products on crypto rails, while differing over how much of the market that represents.