Tech • AI • Robotics
Another AI Safety Kerfuffle, Markiplier’s Stake in GoPro, SpaceX Getting Into Turbines | Diet TBPN
9.4/10TL;DR
A fresh dispute over OpenAI’s use of more opaque reasoning methods has reignited concerns about AI oversight, while separate attention has turned to Markiplier’s GoPro investment and a bizarre fraud case built around a fake NFL identity.
KEY POINTS
OpenAI chain-of-thought dispute
A report from The Information said OpenAI is quietly using more efficient architectures that do not expose a model’s reasoning in plain language, reviving fears about so-called unmonitorable AI. Critics argue that if labs move away from visible chain-of-thought, they lose an early warning system for deception, manipulation and other signs of misalignment.
Safety concerns center on opaque “neuralese”
The debate focused on internal representations sometimes described as “neuralese”, where models pass information through vectors rather than human-readable text. Researchers worried that translating those internal states back into words may miss critical details, meaning a system could hide risky intent in signals that do not survive conversion into tokens.
OpenAI pushed back on the reporting
Jacob Pachocki, OpenAI’s research director, said the reporting was confused and warned against “a race to unmonitorability.” He said the company has tried to preserve and use chain-of-thought monitoring since its first reasoning models, while also acknowledging that current monitoring remains brittle and is getting harder as systems grow more capable.
Broader concern: chain-of-thought may not scale
Even some observers sympathetic to AI safety said visible reasoning alone is an unsound foundation for oversight. The concern is practical as much as technical: future agents could act at very high speed, replicate themselves, collaborate in large numbers and generate logs too vast for humans to meaningfully review in real time.
Cybersecurity warning for AI infrastructure
Ilya Sutskever added a separate warning, arguing that neoclouds have limited cybersecurity and could become prime targets if autonomous agents ever try to seize more compute. The implication is that AI safety increasingly depends not just on model behavior, but on the resilience of the cloud infrastructure running advanced systems.
Markiplier’s GoPro bet drew scrutiny
Online creator and filmmaker Mark Fishbach, known as Markiplier, came under scrutiny after posting a sponsored review of GoPro’s Mission One Pro while also holding a stake in the company. The camera had attracted attention for offering interchangeable lenses and what supporters described as cinema-style output at roughly $700, compared with far costlier traditional rigs.
Timeline suggests product enthusiasm preceded promotion
The sequence appears to have started with genuine enthusiasm for the camera during film work, followed by stock purchases and then the sponsored review. A later correction to public discussion said his stake had been accumulated over time and that his voting power was only about 0.5%, far below the level needed to influence company decisions.
GoPro acquisition changed the story
Days after the sponsored review, GoPro agreed to be acquired by Starman Optical in a deal valued at about $285 million in cash, with shareholders keeping 10% of the combined company. The transaction also preserves the NASDAQ listing, giving the acquirer a public-market vehicle while folding the camera brand into a wider consumer-electronics portfolio.
AI angle complicated the backlash
Some criticism stemmed from Starman’s separate photonics push tied to high-speed optical networking for AI data centers. That created confusion around whether GoPro itself was pivoting to AI, when the more direct strategic logic appeared to be brand fit, distribution synergies and the value of an existing public listing.
Fake 49ers player case exposed social-media fraud risks
Federal authorities said a 35-year-old man allegedly posed as a San Francisco 49ers player and scammed 26 women out of more than $1.3 million through romance and investment schemes. Investigators said he used fake social-media accounts, fabricated bank statements, luxury-car imagery and a made-up football résumé, highlighting how easily online identity claims can be dressed up with plausible digital evidence.
Industrial bottlenecks in the AI buildout are widening
On the infrastructure side, Elon Musk said SpaceX plans to make turbine blades and vanes used in natural-gas power systems, aiming to cut lead times for new generation capacity by as much as 18 months. The move rattled specialist manufacturers because those parts are a bottleneck for electricity supply needed by AI data centers, though analysts noted the manufacturing process is technically difficult and may remain focused on internal use.
CONCLUSION
The common thread across these stories is that AI’s next phase is no longer just about smarter models. It is increasingly about transparency, infrastructure control, capital flows and the credibility of information in a system moving faster than its safeguards.
Explain this