A few items regarding AI's development caught my attention over the last couple of days and should be on your radar:
1. LLMs are not immune from spreading malware and stealing your data. Researchers from Cornell Tech have created a proof of concept that works to do just that. Here's a video showing how it works: https://lnkd.in/gxaQPjDi
2. According to Cohere's President, Martin Kon, Cohere is developing a Generative AI model and toolset designed specifically for enterprise needs, as opposed to consumer-focused chatbots. Seems like a technical person's speak about basic communication models since the output of Cohere still revolves around communication and clarity of communication. However, it's an interesting angle: https://lnkd.in/gKFkMsfW
3. Anthropic released its Claude 3 models today - its most capable model being Opus. Opus is stated to surpasses GPT4 and Geminii Ultra in quite a few benchmark tests including topping the leaderboard on academic benchmarks like GSM-8k for mathematical reasoning and MMLU for expert-level knowledge. It can look at new data, and not rely on data already in it's model to make new analysis and forecasts. Opus can analyze new data without relying solely on its pre-existing model, potentially simplifying a wide range of white-collar tasks to the extent that in the near future, your job may be finding new hobbies. Here's a demo: https://lnkd.in/ggvDaNCk And here’s the Anthropic blog post: https://lnkd.in/gTG7SMZG
4. This Thursday's, The Pandora's Bot Newsletter takes a sharp turn into the job market's unexpected undercurrents, from my perspective. It's the stuff we're thinking (you know you are) but no one is willing to voice. Subscribe here to get it first: https://lnkd.in/gSjMt5aD