Frank Downing
@downingARK
Director of Research, Next Generation Internet @ARKinvest focusing on cloud computing and cryptoassets • Disclosure: http://arkinv.st/2rxmMRG
Yes. I’ve heard “browsers are the ultimate distribution for agents”. I think this is backwards. Agents will use browsers as a tool, among others, to accomplish tasks. You will think of the agent as the product, not the browser. You pay for ChatGPT Agent, which spins up…
websites replace most desktop applications (last 2 decades) then agents replace most websites (next 2 decades)
Grok 3 represented a significant leap forward in performance in February, and now, only 5 months later, the @xai team seems to have done it again. Grok 4 leapfrogs the competition on reasoning benchmarks like GPQA:
Good data points on the importance of "context engineering": Input tokens may be cheaper than output tokens, but context heavy tasks (like coding) can require 300-400x more input tokens of context than output tokens, making context 98% of total LLM usage costs. Latency also…
When you query AI, it gathers relevant information to answer you. But, how much information does the model need? Conversations with practitioners revealed the their intuition : the input was ~20x larger than the output. But my experiments with Gemini tool command line…
After years of reporting 750,000 deployed robots, Amazon has announced the deployment of their millionth.
Two reinforcing flywheels are accelerating AI IDE (integrated development environment) adoption: 1. Model-product flywheel: As foundation models improve (reasoning, planning, tool use), AI IDEs built on top of them instantly get better, driving more product usage. As AI IDEs…
In the 6 months between the general release of @OpenAI's o1 model and the price cuts to o3 last week, the cost of inferencing reasoning models has fallen at a 98% annualized rate. Read more about what we believe the impact of this will be in this week's newsletter:
x.com/i/article/1934…
The ARK Venture Fund is happy to announce that it invested in @neuralink in its Series E funding round! Click to view current holdings and info (subject to change): ark-funds.com/portfolio
Consensus amongst AI leaders is drifting towards the former, while history favors the latter (6/7-day work week shortening to a 5-day work week)
If AI progress continues advancing, are we more likely to see: 20% of knowledge workers unemployed or 100% of knowledge workers working 20% fewer hours?
If AI progress continues advancing, are we more likely to see: 20% of knowledge workers unemployed or 100% of knowledge workers working 20% fewer hours?
Lots of talk about cost declines driven by DeepSeek in China, but underrated is that Chinese co's have pretty firmely taken open source AI lead in terms of absolute performance. 3/4 of the top open soruce models are from Chinese companies 3/3 if you don't count Nvidia's fine…
