@bryanrbeal There is a similar storyline in public transit, especially recently implemented light rail lines. There's a tendency to underestimate the impact of capital refresh programs that will be required.
It also misses the point that the rate of progress on human task efficiency is relatively static. AI continues to show Moore’s Law like improvements in power and cost. Even if humans were currently more cost efficient at particular tasks, AI would likely surpass them in the future.
@b2hcraig Understood - But to the point of the principles you advocate for, just maintaining form probably gets you 90% of the way there, even with a sub-optimal bar.
A more nuanced and accurate story would be that enterprises are adjusting their governance structures (usage monitoring, model selections) as they move from subscription to consumption based pricing. But that would require real journalism and knowledge of the underlying subject matter. Agree with you that most firms haven't even scratched the surface on investments and potential benefits.
@GaryMarcus I am hopeful that GenAI will continue to improve to the point where it does make a significant dent on cancer. In the meantime, there are many ways that, today, creative use of GenAI can make researchers and the entire research process more effective.
@b2hcraig Great form Craig. Unfortunate that many don't understand the virtues of this approach to training. Instead of being open minded and interested in learning, they immediately hurl insults or summarily dismiss the approach.
No mention of the name of the consultant or the client. The underlying story makes little sense. I feel silly debunking it but I think the following is worth considering. Could something of this magnitude happen with the only leak being an unnamed consultant? Does Anthropic not have any guard rails that would trigger early alarms for this level of excessive use? How many public companies could take this level of hit to earnings without needing to disclose it? Sadly, as others have mentioned, this will become an accepted fact, eagerly spread by those with an ax to grind.
As always Ethan, a level headed nuanced take. The general suspicion regarding AI and the rewarding of hysterical headlines with clicks favors the negative press we are seeing. To your point, the answer is not to stop using AI for coding but to refine the practices associated with it.
Agreed Bret. There has been a clear shift from "we need to be using AI" to "are we getting appropriate returns". One side of me says that enterprises need to make serious investments in adding the right tooling, measurements and processes to deal with this issue. The other side wonders whether this problem will be solved by stronger (and more cost effective) models, better vendor supplied tooling and improved practices. We saw similar challenges with earlier models being limited on training data, context windows and connectivity. Enterprises tried to work their way around this with custom capabilities. In the end, the big players solved these issues with their own advances.
So true Craig. Seneca was saying this 2000 years ago. "It is not that we have a short time to live, but that we waste a lot of it. Life is long enough, and a sufficiently generous amount has been given to us for the highest achievements if it were all well invested." — On the Shortness of Life
@BenjaminBadejo Does this support any open source models running on device? Do you have a sense of the operating costs of this using the OpenAI models? Appreciate all the innovative work you are doing here for the community.