What I Read This Week…

U.S. consumer spending may be weakening, Meta released the first open-source model that is competitive with GPT-4o and Claude 3.5 Sonnet, and new blood test can detect Alzheimer’s with 90% accuracy

Chamath Palihapitiya

Jul 28, 2024

Watch All-In E189

Read our Deep Dive on U.S. Defense

Caught My Eye…

Consumer spending in the U.S. is showing signs of weakening, potentially signaling broader economic deceleration. Recent earnings reports from major companies like UPS indicate softening demand, while the consumer sentiment index has fallen to an 8-month low. With consumption representing about 70% of U.S. GDP, weakening consumer spending could negatively impact U.S. economic growth. Analysts attribute the spending slowdown to several factors, including depletion of pandemic-era savings, increasing credit card debt reaching limits, and a slowing labor market.
Meta released Llama 3.1 405B this week, an open-source model trained on over 15 trillion tokens using more than 16,000 H100 GPUs. This is the first open-source foundational model that is competitive with leading closed-source models like GPT-4o and Claude 3.5 Sonnet. As companies building these foundational models exhaust new sources of internet data to train on, the performance gap between open and closed-source models continues to narrow. Developers can now build generative AI applications using open-source models that offer comparable performance to the top closed-source models, while benefiting from the greater customizability and transparency inherent to open-source models.
Scientists have developed a blood test that can detect Alzheimer's disease with 90% accuracy, outperforming dementia specialists using traditional methods by about 20%. How does this new blood test work? It uses mass spectrometry to measure the levels of specific proteins linked to Alzheimer's in a blood sample, including ptau-217, a protein that forms abnormal clusters of tissue in the brains of Alzheimer's patients, as well as various forms of amyloid protein, which accumulate as harmful deposits in affected brains. An algorithm then analyzes these measurements to calculate the likelihood of Alzheimer's disease. This approach offers a less invasive, simpler, and more cost-effective method for diagnosing Alzheimer's compared to current practices.