Skip to content

Guest Talk with Daniel Bashir | A Survey of AI for the Protocol-Minded

TalkDaniel Bashir
researcherpractitioner
Guest Talk with Daniel Bashir | A Survey of AI for the Protocol-Minded
Open resource →

Daniel Bashir, ML engineer at AWS and host of The Gradient podcast, surveys how protocol thinking applies across AI—from funding and scaling laws (like DeepMind's Chinchilla compute-optimal training protocol) to benchmarks (perplexity vs. François Chollet's ARC benchmark for measuring sample-efficient generalization) to regulation (the EU AI Act's risk-tiered categorization). He argues that nearly every layer of AI development—research, engineering, deployment, and governance—can be reframed as a set of competing or evolving protocols, using examples like Sarah Hooker's 'hardware lottery' and Tolga Bolukbasi-style debates over fair/responsible AI operationalization.

Topics

benchmarks-as-protocols-perplexity-arc-vibe-based-evaluationchollets-formal-definition-of-intelligence-and-generalization-hierarchyeu-ai-act-risk-based-regulatory-tiersgovernanceguest-talksprotocolsresearchresponsible-ai-operationalizationscaling-laws-and-the-bitter-lessontechnology

Related resources

Talk

Guest Talk with Cory Levinson | Who Writes the Rules of Climate Protocols?

Cory Levinson traces the history of carbon markets from the Kyoto Protocol through the Paris Agreement's still-unoperationalized Article 6.4, contrasting the fragmented voluntary carbon market (Verra, Gold Standard, ACR) with compliance regimes, and recounts how Toucan's CO2 (CLA) tokenization protocol exploited Verra's retirement mechanism until Verra froze bridging in 2022. He then frames climate protocols into three layers—financial, scientific (MRV/methodologies), and semantic (ontologies/RDF)—arguing that semantic protocols using relational data models like RDF and the semantic web could let disparate scientific methodologies build on each other rather than scaling in isolation.

Cory Levinson

Talk

Psychological Safety I & II

Timber Stinson-Schroff argues that AI's workplace effects should be studied through occupational health frameworks rather than existential-risk AI safety, proposing 'cognitive ergonomics' as a better term than 'psychological safety' since it emphasizes chronic, long-term pathologies (like carpal tunnel from typewriters) over acute incidents. Drawing on the safety-one/safety-two distinction from traditional safety literature, Stinson-Schroff advocates crowdsourcing a database of AI usage hazards and pathologies—such as the 'yes man problem' from fast LLM responses and speech pattern changes—toward an eventual diagnostic manual, while cautioning against liability blind spots given the slow, second-order mechanisms by which these harms emerge.

Timber Stinson-Schroff

Talk

Trust Experience Glitches: AI Agents, Delegated Autonomy, and Legitimacy

Helena Rong (Assistant Professor at NYU Shanghai, director of the Protopolis Lab) presents a taxonomy of 'trust experience glitches'—six failure modes (agency assertion, epistemic camouflage, delegation ambiguity, opacity at scale, cascading autonomy, revocation failure) drawn from a 167-incident dataset of agentic AI failures. She argues these ruptures expose a 'consequence reception problem' where autonomous agents lack a stable body to bear accountability, and proposes that governance must shift from ex-post legal remedies toward upstream protocol primitives (capability-based permissions, provenance tracking, action-class legibility, circuit breakers, revocability) as the 'traffic lights' of the agentic web.</summary>

Helena Rong