Does What We Do Even Matter? - WAN Show January 31, 2025

Episode Summary

Executive Summary: The episode centers on two big shifts: whether GPU reviews still matter when flagship cards sell out instantly, and how DeepSeek is reshaping AI economics by making local LLMs practical. The hosts also cover open-source and platform changes, including Pebble’s revival, SteamOS expansion, Sony dropping PSN requirements on some PC games, and AI-poisoning tools, while repeatedly emphasizing that utility, not hype, should guide tech adoption.

Main Topics: Do GPU reviews still matter? (Priority: 5/5): The hosts question the value of spending hundreds of lab hours on GPU reviews when high-end cards sell out immediately and many buyers never watch reviews. They debate whether reviews still influence the market or mainly serve a smaller, highly engaged audience. DeepSeek and the local LLM surge (Priority: 5/5): DeepSeek’s efficiency and open-source release are presented as a major catalyst for local AI experimentation. The discussion focuses on how it lowers hardware barriers, changes consumer behavior, and may pressure OpenAI and Nvidia economically. GPU scarcity, scalping, and pricing power (Priority: 4/5): They discuss 50-series launch shortages, bots scalping stock, and inflated resale prices as evidence that Nvidia faces little incentive to cut prices in a low-competition market. Open-source and self-hosted software revival (Priority: 4/5): Pebble’s return, SteamOS support outside the Steam Deck, and local LLM deployment all illustrate a broader preference for open, self-hostable systems that reduce dependence on cloud services and subscriptions. AI poisoning and anti-scraping defenses (Priority: 4/5): The episode reviews multiple ways creators can make scraping harder or less valuable, including malformed subtitles, Glaze/Nightshade, and tar-pitting tools, while noting these are defensive, not decisive, measures. Platform policy shifts and gaming access (Priority: 3/5): Sony removing PSN requirements for some single-player PC games is framed as a positive correction after backlash over region restrictions. The hosts also note Xbox/PlayStation crossovers like Forza Horizon 5 on PS5. Odds and ends: merch, schools, and nostalgia (Priority: 2/5): The show includes lighter tangents on school sports infrastructure differences, old game nostalgia, cybernetic futures, keyboard issues, and audience Q&A about everything from warranties to morning routines.

Key Arguments: GPU review audiences are much smaller than the total gaming market, so reviews may not materially change purchasing behavior for most buyers. Even if review viewership is limited, lab work remains valuable because it provides important information and serves the people who do watch thoroughly. DeepSeek appears to have accelerated local LLM adoption by making capable models cheaper, more efficient, and easier to run on consumer hardware. Open-source AI reduces dependence on cloud providers and may undermine the pricing power of incumbents like OpenAI. The real battleground for AI is economic: if models can be run locally, expensive subscription-based services face stronger pressure. Nvidia’s market power and lack of competition let it maintain high prices even amid criticism and scalping. Many AI-poisoning or anti-scraping tools are only partial defenses, but they can still help individual artists and publishers protect specific styles or data. SteamOS becoming more widely compatible could meaningfully improve handheld and living-room PC gaming experiences. Sony’s PSN requirement rollback shows that publisher backlash can force policy reversals when legitimate customers are blocked. Open-source, self-hosted tools are increasingly seen as more trustworthy and flexible than cloud-only alternatives.

Data Points: Nvidia stock drop: 17% - The market reaction on Monday January 17 after DeepSeek-related news Nvidia market cap loss: roughly $600 billion - Estimated value erased during the tech selloff DeepSeek training cost: less than $6 million USD - Reported cost to train DeepSeek, though the hosts note the figure is not fully verified Total addressable audience for GPU reviews: 4–6 million people - Estimated overlapping audience that actually watches hardware review videos PC gamer population: 1–2 billion - Used to show how small the review-watching audience is relative to the overall market Review retention: below average; under 50% by the Blackwell/neural rendering section - The hosts cite low retention on their RTX 5090 review RTX 5090 price: $2,000 - Discussed as a high-end card that sold out immediately RTX 5090 resale price: $5,800 - A Zotac 5090 was found listed on Marketplace later the same day RTX 3090 launch MSRP: $1,499 - Used to calculate long-term value retention of older GPUs RTX 3090 current resale: $800–$1,000 USD - Observed used prices due largely to 24GB VRAM demand RTX 3090 value retention period: about 50 months - From launch in September 2020 to the present discussion Estimated monthly depreciation on a 3090: about $12/month - Calculated from launch price versus current resale GeForce Now Ultimate price: $20/month - Compared against owning a high-end GPU GeForce Now session limit: 8-hour sessions - Mentioned as a practical limitation for some users DeepSeek forks on GitHub: 6400+ - Used to emphasize the model’s open-source spread Mac mini RAM ceiling: 32GB on M4, 64GB on M4 Pro - Discussed as a practical local LLM platform Mac Studio RAM ceiling: up to 192GB - Mentioned as a stronger option for local LLM workloads Pebble reboot timing: acquired by Google in 2021; Fitbit acquisition in 2016 - Timeline for Pebble OS/open-source revival discussion School stadium example: Allen High School has multiple baseball/softball fields and a large stadium - Used to contrast U.S. school sports infrastructure with Canada PSN-enabled Sony PC games: 4 games - Spider-Man 2, The Last of Us Part 2, God of War Ragnarök, Horizon Zero Dawn Remastered

Pivotal Quotes: "does what we're doing even make a difference" — Linus: Introduces the main topic about whether GPU reviews still matter "we spend how many person hours went into something like our 50 90 and 50 80 reviews" — Linus: Argument about the labor cost behind flagship GPU reviews "what it really just boils down to for me is i'm stoked" — Linus: Pebble revival discussion, expressing enthusiasm for the open-source reboot

Implications: The episode suggests tech buying is shifting toward self-hosted, open, and utility-driven solutions. Review media still matters, but mostly for a smaller informed audience. DeepSeek may accelerate local AI adoption and pressure cloud AI economics.