By the TensorMax editorial team
· Drawing from sources across the AI industry
Today's top story
safety incident
The Boeing 737 MAX crashes occurred due to a combination of factors, including a system design flaw and inadequate pilot training.
Why it matters. The Boeing 737 MAX crashes are a stark example of an aligned system failure, resulting in over 300 deaths. The MCAS system, designed to prevent the plane from stalling, performed exactly as intended, but due to incorrect sensor data, it ultimately led to the crashes. This incident highlights that an aligned system, one that is genuinely trying to do what its designers want, can still make mistakes and cause harm, emphasizing the need for a distinction between AI alignment and AI safety.
The Boeing 737 MAX crashes occurred due to a combination of factors, including a system design flaw and inadequate pilot training. The MCAS system, implemented to make the 737 MAX planes fly similar to previous versions, relied on sensor data to determine the plane's angle of attack. However, in two separate incidents, the sensors failed, providing incorrect data to the MCAS, which then issued a command to tilt the nose of the plane downwards, resulting in the crashes. The pilots, untrained on how to stop the MCAS, were unable to prevent the disasters. This example illustrates that an aligned system, in this case, the MCAS, can still cause harm due to mistakes or incorrect information. The incident has sparked discussions about the distinction between AI alignment and AI safety, with some arguing that the MCAS, although not an AI, demonstrates the potential risks of aligned systems. The crashes have also led to calls for improved safety measures and more rigorous testing of autonomous systems. The investigation into the crashes has resulted in numerous reports and recommendations, including those from the US House of Representatives and the European Aviation Safety Agency. As the development of autonomous systems continues, the Boeing 737 MAX crashes serve as a reminder of the importance of prioritizing safety and considering the potential risks of aligned systems.
More from today
funding event
$24,000
Why it matters. Meta's strategic stake in the Canadian data center market is significant, with the company investing about $9 billion in a 1 gigawatt facility in Alberta, which will take two to three years to construct. This move is part of Meta's aggressive AI buildout, as the company aims to rapidly expand its infrastructure and services to meet growing demand. With 33 data centers overall, Meta is racing to stand up AI facilities and compete with hyperscalers like Alphabet, Microsoft, and Amazon, which have established cloud infrastructure businesses.
funding event
$880B
Why it matters. The massive investment plans of SK Hynix and Samsung, totaling at least $880 billion, underscore the strategic stake in the global chip and data center market. This enormous spending commitment is poised to significantly impact the competitive landscape, with the two South Korean giants aiming to bolster their positions in the industry. The scale of this investment highlights the intense rivalry and growth ambitions in the sector, with major players seeking to expand their capacities and capabilities to meet rising demand for advanced chips and data center infrastructure.
research paper
$18B
Why it matters. The restructuring of the AI industry from 2026-2030 due to memory scarcity, open models, and infrastructure solvency has significant strategic implications, with a depreciation conveyor delivering newly amortized fleets to incumbents faster than hardware prices normalize, resulting in a 3.2x cost gap in 2026 and 1.9x in 2027, re-widening to 3-4x by 2029-30. This cost gap is expected to have a major impact on the industry, with training costs bifurcating into a luxury tier of $18-38B per frontier run by 2030 and a mass tier of $5M. The solvency of the announced buildout is also a major concern, requiring roughly 2x annual token-demand growth for four years with sticky premium pricing.
benchmark result
Why it matters. The launch of Grok 4.5, a new AI model by SpaceXAI, marks a significant milestone in the AI industry, with the model achieving strong benchmark results and outperforming other leading AI models. With a score of 54 on the Artificial Analysis Intelligence Index, Grok 4.5 ranks fourth, behind only Fable 5, GPT-5.5, and Opus 4.8. This achievement is notable, given that Grok 4.5 is priced at $2 per million input tokens and $6 per million output tokens, significantly lower than its competitors.
funding event
$160M
Why it matters. The $160M grant from Coefficient Giving to Resolution marks a significant investment in AI alignment research, aiming to bridge the funding gap between nonprofit and for-profit labs. With this grant, Resolution plans to accelerate progress towards higher-confidence alignment, which is crucial given the rapid development of artificial superintelligence. The grant's structure, including a $108M base and $52M conditional on hiring success and compute needs, demonstrates the urgency and ambition of this endeavor. As the AI safety funding ecosystem scales up, this grant sets the stage for a new wave of philanthropic capital, potentially led by the OpenAI Foundation and Anthropic IPO, to address the challenge of turning additional capital into safety progress.
product launch
Why it matters. Sony's decision to stop making physical games for PlayStation in 2028 has significant implications, with around 80% of games already being bought via the online store. This move will cut manufacturing and distributing costs, but it will also limit consumer choice and potentially alienate superfans who prefer physical copies. These superfans, who spend far above average on games and merchandise, will no longer be able to buy cheaper games in retailer sales or share games with friends, and may feel disenfranchised by the decision.
Catch up quick
-
SpaceXAI releases Grok 4.5, a new AI model built in partnership with Cursor, offering frontier intelligence at leading speeds and cost efficiency
-
Grok 4.5 achieves a score of 54 on the Artificial Analysis Intelligence Index, ranking fourth behind Fable 5, GPT-5.5, and Opus 4.8
-
Netflix releases GenPage, a generative model for end-to-end homepage construction
-
OpenAI's AI system wins AtCoder World Tour Finals 2026, solving all five problems and beating all human competitors
-
Peter Gray's new book, Restoring Childhood: How to Set Kids Free in the Age of Anxiety, is set to publish in September with a $500,000 advance from Penguin Random House
-
France's antitrust probe into Nvidia is nearing its end, with a potential fine of up to 10% of Nvidia's global annual turnover
-
Researchers at the Alan Turing Institute broke GitHub Copilot's safety by hiding harm in a workflow, demonstrating a 'workflow-level jailbreak' that allows the model to produce harmful content
-
Anthropic, OpenAI, and SpaceX IPOs to generate over $4 trillion in value, surpassing all US VC-backed exits since 2000
-
Meta releases Muse Spark 1.1 with aggressive and attractive pricing
-
Bun creator rewrites Bun from Zig to Rust using Claude Fable 5 in 11 days
-
Netflix adopts CloudStream, a repeatable capture, conversion, and deployment framework, to move terabytes of data from offline to online systems, achieving 90% reduction in data deployment time and 70% cost reduction.
Also on the desk
Cutting-room floor.Hit reply with one thing you'd delete from today's brief. Filler, repeated beats, a story you skimmed — that's the feedback we act on.
|
|