Sign in Subscribe
3 min read

Week 49, 2024

Subscribe

-

-

Signals

-

  1. Companies need useful tools, not cutting-edge models
  • BBVA's experience tells the story - even with 3,000 custom GPTs saving employees hours weekly, they can't integrate AI deeply with core banking systems.
  • Reuters faced the same reality check, choosing to split tasks between different AI models based on practical strengths rather than chasing the most advanced model.
  • Inflection AI's pivot makes perfect sense in this context - they abandoned the frontier AI race to focus on enterprise tools because that's where real demand lies.
  • Alibaba's Marco-o1 completes the picture, deliberately targeting practical business problems instead of chasing benchmark scores.
  • Together, these moves reveal how enterprise AI is entering a new phase: less about impressive capabilities, more about fitting into existing workflows and solving specific problems.
  1. Tech's boldest promises still run on human labor.
  • The pattern keeps repeating: every "fully autonomous" solution seems to have humans quietly working behind the scenes.
  • Tesla just revealed they'll need remote operators in VR headsets to control their supposedly self-driving robotaxis - a striking admission from a company that's been promising true autonomy for years.
  • But it's not just Tesla. Uber's latest venture tells an even more revealing story.
  • They're entering the AI training business by offering their gig workers as data labelers, paying as little as $2.37 per task in some regions.
  • From Tesla's remote drivers to Uber's data labelers to customer service chatbots that quietly hand off to human agents, there's a growing gap between automation marketing and operational reality.
  1. Open platforms struggle with data control while closed ones battle misconceptions.
  • Bluesky's commitment to openness accidentally created a data privacy nightmare.
  • Their public API meant anyone could scrape a million posts for AI training, forcing them to admit they can't actually enforce user consent outside their system.
  • Meanwhile, Microsoft faced the opposite problem - a viral panic about Office using documents for AI training that turned out to be completely false, but spread anyway because people now expect the worst about their data.
  • The irony is perfect: the open platform can't protect user data even when it wants to, while the closed platform can't convince users their data is safe even when it is.
  1. Everyone's building backdoors to Big Tech dependency
  • Smaller tech companies are making coordinated moves to break free from platform gatekeepers.
  • Anthropic's new Model Context Protocol isn't just another API - it's a clever way to let any AI assistant talk to any app, potentially bypassing the AI giants' control of data flows.
  • Perplexity's $50 voice device concept, despite following Humane and Rabbit's troubled path, shows the same instinct: build direct user relationships without Big Tech intermediaries.
  • Roblox just demonstrated the most direct approach - offering users 25% more virtual currency for bypassing Apple and Google's stores.
  • It's particularly bold considering they're not just avoiding the 30% fee, they're explicitly showing users the cost of platform dependency.
  • Tech companies have learned the platform dependency lesson and are actively building escape hatches - whether through protocols, hardware, or economic incentives.
  1. Advanced chip manufacturing proves immune to fast-tracking, even with billions in funding and political pressure.
  • TSMC's Arizona timeline keeps slipping.
  • Their confident promise of 2nm chips by 2028 has quietly shifted to 2030, and even Taiwan's government admits that might be optimistic.
  • Intel's story completes the picture: despite $7.86B in CHIPS Act funding, their Ohio timeline just moved from 2025 to "end of decade."
  • Despite the political will and massive funding, they're hitting the same walls as TSMC.
  • These delays present reality checks on the entire concept of rapidly transplanting advanced manufacturing.
  • You can't accelerate decades of manufacturing expertise with money alone.
  • Taiwan's advantage isn't just about factories - it's about an ecosystem of knowledge built over generations.

-

-

Rundown

-

AI

Nvidia launches AI voice maker Fugato.
Amazon building video AI Olympus.
OpenAI Sora leaked by artists.
OpenAI trademarks o1.
OpenAI sells $1.5B shares to SoftBank.
X investors given 25% of xAI.
LinkedIn post mostly AI-written.
Zoom rebranded with AI focus.

Robotics

Neuralink starts robotic arm trials.

Chips / Infrastructure

Meta laying 40,000km sea cable.
Intel gets chip funding w/ conditions attached.
TSMC Arizona 2nm chips delayed.
Xiaomi plans 3nm chip production.
Huawei poaching Western chip experts

Devices / Hardware

Apple working on foldable iPhone.
Cook makes third China trip this year.
Huawei launches phone with own OS.
Snap AR glasses chase tech over style.

Content / Entertainment

Apple-Real Madrid discuss VR streaming.
Spotify blocks third-party data access.

Social Media

Bluesky faces verification issue.
Reddit pushes global expansion.
TikTok bans beauty filters for under-18s.
Australia bans social media for under-16s.

EV / Autonomous

Pony.AI raises $260M IPO.
Tesla-Rivian settle trade secrets lawsuit.
CA readies EV rebate backup plan.

Space

Starlink direct-to-cell approved.
SpaceX plans 25 launches in '25.

Zuck meets Trump at Mar-a-Lago.
Hoffman worries Trump retribution.
Senate pushes AI training data disclosure.
Canadian media sues OpenAI.

Etc

Entry-level coding jobs vanish.
MicroStrategy surges on bitcoin.
Google Maps led to three lives lost.

-

Subscribe to Big Tech Digest