May 10, 2024, 6:02 PM ETInterview, Fireside Chat, Podcast

Sam Altman: Getting Fired (and Re-Hired) by OpenAI, Agents, AI Copyright issues

GPT-5 Release: No confirmed date; likely a "continuous improvement" model rather than a discrete version release, with potential for a different release mechanism than previous iterations.
Accessibility Strategy: OpenAI aims to make advanced AI (GPT-4 level) available to free users despite high costs, prioritizing widespread utility over revenue in the short term.
Open vs. Closed Source: OpenAI maintains a closed-source strategy for frontier models to ensure safety and speed, though it acknowledges a significant role for open-source models, particularly for on-device applications (e.g., running on smartphones).
Model Architecture Shift: The industry is moving toward specialized models for specific tasks (video, biology) rather than a single "one model to rule them all," though OpenAI envisions generalized reasoning eventually connecting these domains.
Sora Capabilities: Sora (video generation) uses a custom architecture distinct from language models, not starting with a language foundation.

Hardware Constraints: The primary bottleneck for the industry remains NVIDIA chip throughput, energy supply, and data center construction logistics.
Device Future: Sam Altman suggests the "next big thing" will be a new form factor rather than a cheaper phone; he believes the iPhone sets an exceptionally high bar and that new interactions (voice, visual) must be significantly different to justify adoption.
Apple Partnership: Altman and Johnny Ive are discussing new device concepts, though no specific product details were confirmed.
Compute Investment: Reports suggest Altman is exploring massive infrastructure projects (e.g., a "chip project" potentially targeting $7 trillion scale), though exact figures and structures remain speculative.
Pricing Model: Altman anticipates a future where AI intelligence becomes "cheap to meter" and "fast enough to feel instantaneous," unlocking new use cases.

Regulatory Approach: Altman advocates for international oversight of frontier AI models (similar to nuclear or bio-regulation) for systems capable of causing significant global harm, rather than detailed statutory audits of code by individual nations.
Startup Concerns: He supports a threshold-based regulation (e.g., models requiring >$10B in compute training) to avoid stifling startups while ensuring safety for the most powerful systems.
Board Dynamics: The November 2023 firing resulted from a culture clash between non-profit board members prioritizing safety and Altman's focus on speed and commercial scale; the board cited safety concerns over rapid advancement.
Equity Structure: Altman confirmed he eschewed personal equity in OpenAI to maintain a "disinterested" non-profit board, a decision he later admitted he would have reconsidered to avoid perception issues regarding motivation.

Artistic Rights: Altman argues that AI training on existing data is distinct from human learning but acknowledges the need for new economic models, particularly regarding style mimicry (e.g., generating in the style of Taylor Swift).
Current Policy: OpenAI does not currently allow generating music or reproducing specific artist styles without consent; they are exploring opt-in/opt-out mechanisms and economic compensation models.
Legal Stance: OpenAI maintains a "fair use" position for training data but acknowledges the law is evolving and anticipates a shift from training data debates to inference-time behavior debates.
IP Protection: The company actively blocks the generation of copyrighted characters (e.g., Darth Vader) but allows stylistic derivatives (e.g., "Sith bulldog") which exist in the cultural sphere.

AGI Vision: OpenAI's explicit mission is to ensure Artificial General Intelligence (AGI) benefits all humanity, a goal Altman views as both inevitable and beneficial.
AGI Safety: Altman expresses low concern regarding current GPT-4 level models causing catastrophic harm but warns that future, recursive self-improving systems will require rigorous international safety testing.
UBI Evolution: Altman suggests the conversation around Universal Basic Income (UBI) may shift toward "Universal Basic Compute," where individuals receive a share of AI productivity or compute resources.
Scientific Discovery: Altman identifies scientific discovery (e.g., protein folding, drug design) as the most transformative application of AI, emphasizing that reasoning capabilities, not just language generation, are the key enabler.
Healthcare Transformation: He predicts AI will fundamentally transform healthcare through faster drug discovery and personalized medicine.

The Firing: Altman confirmed he was fired on November 30, 2023, via text and phone call from the board; he initially considered doing independent AGI research before returning to lead OpenAI following a team intervention.
Interim CEO: The board hired a different interim CEO (Mira Murati was mentioned in passing as part of the team, but the interim role was filled by a former Scaramucci figure) before Altman's return.
Relationship with Board: Altman respects the board's integrity and commitment to safety but disagrees with their decision-making process regarding the timeline of AGI deployment.
Poker: Altman mentioned a history of playing high-stakes poker with the podcast hosts, including a specific hand where he went all-in.

Apple Innovation: The hosts critique Apple's recent product stagnation (thinner iPads, lack of innovation) and note Tim Cook's recent stock sales despite public assurances of long-term holding.
Google AlphaFold 3: Discussed as a major breakthrough in biology, allowing the prediction of protein-small molecule interactions, though Google has kept the core IP closed for commercial use within its Isomorphic Labs subsidiary.
Open Source Dynamics: The consensus is that while base models may eventually converge in capability, the "scaffolding" (apps, infrastructure, safety) will become the primary source of value, similar to the open-source software ecosystem running on proprietary cloud infrastructure.
Protest Ethics: The hosts debated the morality and legality of campus protests, agreeing that peaceful assembly is protected but that police crackdowns and "safetyism" rhetoric can sometimes be excessive or counterproductive.