Concerned mom uses ChatGPT to solve a diagnosis 17 doctors couldn't make

Welcome to The Dispatch! We are the newsletter that keeps you informed about AI. Each weekday, we scour the web to aggregate the many stories related to artificial intelligence; we pass along the news, useful resources, tools or services, technical analysis and exciting developments in open source. Even if you aren’t an engineer, we’ll keep you in touch with what’s going on under the hood in AI.

AI’s past, present and future through the eyes of an historian of science

From University of California Berkeley: Historian of science Cathryn Carson offers perspective and contemplation upon AI's past, present, and future. She argues that AI bringing new benefits and risks is not unlike previous historical breakthroughs, and advocates for systemic changes to address issues like data privacy and corporate tech power:

“We have to acknowledge that AI and machine learning and automated decision systems are built on data that are extracted from people and communities, usually without their full knowledge, sometimes — maybe often — against their interests. Those systems are then used to govern those folks and shape their decisions and outcomes and opportunities and serve other people who make the systems and run them.”

More Details:

• Carson argues that the stakes and downstream consequences of AI are more comparable to the impacts of plastics and fossil fuels than they are to (the commonly analogized) nuclear weapons. Like plastics and fossil fuels, AI is being developed primarily by corporations and embedded in markets, not just nation-states.

• She sees familiar societal dynamics around AI like hype, demands for regulation, and downstream consequences. She argues we need to understand how we got here, should we want to change these patterns or look into the future.

• The U.S. currently operate under patch-worked 1970s data protection laws not designed for today's AI data collection. She advocates for expanding beyond small personal choices to an intentioned systemic change in the power dynamics allowing data use against people's and the public’s interests.

Takeaways: The analogy to plastic is thought-provoking. We don’t necessarily agree that the stakes here are similar to the stakes around the refinement of plastics or fossil fuels. We feel they’re higher (but maybe we’re biased). We do whole-heartedly agree that addressing the complex questions surrounding AI's development and governance (including systemic changes around data) will require a collective effort on the part of AI scientists and engineers, social scientists, policymakers, ethicists, impacted communities. Each has an invaluable perspective to contribute.

ChatGPT analyzed the data and suggested tethered cord syndrome as a possible diagnosis. Courtney researched the condition and found a neurosurgeon who confirmed Alex had occult spinal bifida causing a tethered spinal cord. This congenital abnormality stretched his spinal cord during growth, resulting in pain and mobility issues. Alex has since had a successful surgery to untether the cord.

While ChatGPT should not be used to confirm medical diagnoses, it can be used to help unearth them. The Harvard epidemiologist in the article highlights the value of AI-copiloting: "I do think ChatGPT can be a good partner in that diagnostic odyssey. It has read literally the entire internet.” That is not a factually accurate statement, but the sentiment registers.

• AI app CharacterAI is catching up to ChatGPT in the US

• UK won’t cut China out of AI summit over spying scandal

• (Salesforce) New AI usage data shows who’s using AI - and uncovers a population of ‘super-users’

• Another group of writers is suing OpenAI over copyright claims

• AI-generated Drake and The Weeknd track won’t be winning a Grammy

The result? Faster query response and reduced API expenses. According to the authors, adopting Doris has streamlined their analytics stack. With optimizations like LLM parsing and architectural refinements, Tencent has built an intelligent analytics framework on Apache Doris that can interact with natural language rather than SQL. Their learnings provide valuable insight for any organization looking to scale their analytics with open source OLAP.

The LLM then outputs text responses directly, and "signal tokens" that trigger generation in other modalities by corresponding decoders. In testing, the system handles free-form conversational prompts, correctly interpreting inputs in one modality and generating appropriate responses in different requested modalities. This is definitely an ‘early days’ multimodal LLM infusion project that does not yet have ultra-high quality outputs (and there are other multimodal AI systems), but the fact that NExT-GPT offers multi-modal input is a real breakthrough.

• Machine Learning papers explained

• Why Nvidia’s AI supremacy is only temporary

• The real problem with software development