Simple Phrase Improves LLMs, AI Wave for Enterprise SaaS, AI Jobs Clustering, & Disney Uses AI for Loki Image
DeepMind shows how a simple phrase can improve accuracy, a16z General Partner discusses AI's SaaS infusion, 60% of AI jobs are in 10 metro areas, and Disney's AI image causes a stir
DeepMind Shows How A Simple Phrase Improves LLM Accuracy
As more resources pour into generative AI, it’s easy forgetting about the simple things. Like telling a Large Language Model (LLM) to "take a deep breath and work on this problem step-by-step." This friendly nudge scored highest with LLMs, showing just how similarly large language models can think like humans.
As you can see below, other phrases like “break this down” also improved LLM accuracy.
DeepMind’s method behind the madness was Optimization by PROmpting (OPRO), which optimizes tasks defined in natural language instead of formal mathematical definitions.
Aside from leveraging LLMs as optimizers, DeepMind shows how the right instructions, and even a simple phrase, can be a game-changer. Even though we’re working with machines, a human touch is still needed.
SynthAI: The Next Wave of Enterprise Software
Kristina Shen, General Partner at Andreessen Horowitz, recently presented on how AI will impact software over the next several years.
She contrasted how consumers care more about generative AI, while enterprises will likely focus more on synthesizing data (SynthAI), where their troves of data are synthesized into actionable insights.
As to who will win the SynthAI battle between entrenched software companies and AI native startups, that’s up for debate. Shen explains how Zendesk and ZoomInfo have already successfully embedded AI, but this battle has just started.
It was only a few years ago where studios felt that Netflix’s application could be easily duplicated, and that the content they created was the real work. Thus, all of the studios created their own streaming apps: Disney (Disney+), Warner Brothers (Max), Paramount, and Comcast (Peacock). Doing so, they felt, would help them capture more value.
Well, it turns out that just “building a streaming app” may be a little more difficult in hindsight. Netflix has soundly outperformed since those studios launched their own apps.
Could software incumbents be overlooking how difficult AI is? Given how rarely we see incumbents succeed when new technologies are introduced, it’s certainly possible.
AI Job Boom: A Mirage of Decentralization Amidst the Clustering Reality
Despite the democratizing hopes with generative AI, the reality has been more of the same. New data from Brookings shows generative AI jobs are concentrating within the major tech hubs, mirroring the historic winner-take-most dynamics of the digital sector.
Over 60% of these jobs are clustered in just 10 metro areas, with nearly a quarter in the Bay Area alone. Deep talent pools and substantial resources have focused AI activity to these regions, trumping the potential for a more geographically inclusive AI landscape.
Disney’s Loki 2 Announcement Shows Public Disdain with AI
Even a simple tv show reminder can’t escape controversy.
Disney told their audience to “mark their calendar” for season 2 of Loki, and included the image below. Seems innocent enough, until you compare the Roman numerals in the background clock (yellow arrows), specifically how different the V’s look.
That was enough for many to declare that AI was used in making this image, and the replies were quite telling.
While it’s unlikely that these commenters would watch Loki regardless of the image Disney used, these replies show how antagonistic many are towards AI. While it will be impossible putting the AI toothpaste back in the tube, it will also be a lot tougher innovating if the public is against this technology.