Unsupervised Learning: Redpoint's AI Podcast• Dec 18, 2025• 1:18:04Interview

AI Vibe Check: The Actual Bottleneck In Research, SSI’s Mystique, & Spicy 2026 Predictions

From Unsupervised Learning: Redpoint's AI Podcast

Jacob(Host)•Rob(Guest)•Ari(Guest)

Executive Summary

The rate of fundamental improvement in large language models has significantly slowed, indicating a plateau in the current paradigm, which shifts focus towards application-layer value and infrastructure.
Competition among major AI labs is intensifying, with OpenAI's market position seen as weaker than 18 months ago compared to Google's structural advantages, leading to bold predictions of leadership changes.
The AI research landscape is becoming more closed as labs like Google and Meta restrict the publication of meaningful work, turning academic conferences like NeurIPS into PR events rather than forums for open science.
A strong conviction exists that open-source models, potentially from China, could surpass proprietary models by 2026, challenging the dominance of US-based labs and altering the geopolitical landscape of AI.

12 quotes

Concerns Raised

The slowing rate of fundamental LLM improvement could signal a paradigm bottleneck.
OpenAI's high cash burn and weakening competitive position relative to Google.
The trend of closed research at major labs may stifle broader innovation.
US chip controls may be backfiring by accelerating China's domestic AI capabilities.

Opportunities Identified

Trillions of dollars in economic value can be created by applying current-generation AI models.
The plateau in model progress creates stability for a new wave of AI infrastructure companies to emerge.
Open-source models present a significant opportunity to challenge the dominance of closed, proprietary systems.
New research labs ('neo labs') focused on novel paradigms could become the next market leaders if they achieve a true breakthrough.

Key Themes

The AI Model Plateau

The discussion posits that the dramatic, step-function improvements in LLM capabilities seen leading up to GPT-4 have significantly slowed. While progress continues, the rate has flattened, suggesting the current scaling-based paradigm is reaching the mature phase of its S-curve.

This plateau shifts the industry's focus from foundational model breakthroughs to commercial applications and infrastructure. It creates stability that allows a new wave of infrastructure companies to emerge and forces incumbents to compete on product and distribution rather than raw capability.

Intensifying AI Lab Competition

The competitive dynamics between major AI labs are in flux. OpenAI's once-unassailable lead is now challenged by players like Google, which possesses deep talent, vast compute resources, and distribution advantages, leading to speculation about OpenAI's high burn rate and future leadership.

The strategic decisions and relative strength of these labs will define the market structure for the next decade. An IPO, a leadership change, or a strategic pivot by any of the major players could significantly alter the investment and competitive landscape.

The Closing of AI Research

As AI becomes more commercially critical, major corporate labs are increasingly secretive, preventing researchers from publishing their most impactful work. This has changed the nature of academic conferences like NeurIPS, which are now seen as less about scientific exchange and more about PR and recruiting.

This trend centralizes cutting-edge knowledge within a few corporate entities, potentially slowing broader scientific progress and making it harder for startups and academia to compete on the technological frontier.

The Geopolitics of Open Source AI

There is a strong belief that an open-source model will soon outperform all closed-source alternatives. A specific prediction is made that a Chinese open-source model will be the world's best at least once in 2026, suggesting US export controls may be inadvertently accelerating China's domestic AI ecosystem.

The emergence of a top-tier open-source model from China would fundamentally reshape global AI power dynamics, challenge the business models of proprietary AI companies, and complicate the geopolitical strategy of technological containment.

The 'Neo Lab' Dilemma

The viability of new AI research labs ('neo labs') aiming to create the next paradigm is debated. The bull case is that a breakthrough could secure a spot in the next AI oligopoly, but this is countered by skepticism due to the fast-follower effect and the historical pattern of simultaneous independent discovery.

This questions the defensibility of fundamental research breakthroughs in the current environment. It suggests that a technical lead may be short-lived, and that go-to-market strategy, distribution, and capital are equally critical for a new entrant's long-term success.

Get started free

Topics

AI Models Large Language Models (LLMs)GPT-4 OpenAI Google Meta FAIR Sam Altman AI Research NeurIPS Open Source AI US-China Tech Competition Chip Export Controls AI Infrastructure Scaling Laws Venture Capital AI Startups

Processed Apr 15, 2026 yt-dlp + mlx-whisper + Gemini