2,711 karma · joined March 20, 2015
—
But this is never the problem. Claude WILL NOT abstract and WILL NOT use your abstractions. It finds them all “ceremonial” and the idea that you could add something that might seem indirect that actually dramatically reduces the problem space is almost impossible to convey.
You can watch this in action for any API whose design you’re familiar with in a domain you understand well. If you attempt to design the same API with Claude, your will invariably get a mess of flat, insane types and no reuse. I’m talking an array of tuples of maps of set to map type insanity.
What has been helping is a mandatory pass of “Claudisms”, but even then it can only find the problem and never the solution.
It is so frustrating.
The idea that we’re at the point where it’s superseded our ability to tell just makes no sense. I’ll be happy if we can get to a point where I don’t have to tell Claude not to tail every bash command or make a job that writes throughout instead of once at the end. I’ll be happy if “continue this interaction naturally, you are taking over from an independent subagent” works.
But I’m not holding my breath. It’s still really cool that any of this stuff is possible.
I’ve had doctors try to convince me not to pursue medical care, that problems of people close to me were not real and purely psychological, and I’ve personally required emergency surgery due to inaction. In every case there were obvious signs and symptoms.
Doctors are not good at their jobs. In the US, we’ve done a particularly stupid combination of forcing them to incur legal liability and intermediating everything with insurance, both of which impact the care people actually receive.
If everywhere built more housing do you think the price of housing would go up? Where are these people bringing up housing prices coming from?
Look at Japan, look at any metropolitan U.S. city that has actually leaned into it. Europe has had mixed use zones for effectively centuries and is not the dystopia that NIMBYs proclaim will appear in the absence of zoning.
It is unpopular because we subsidize the lives and assets of people who “have things” through zoning policy that they make.
I also let him play at the park on his own occasionally. I will get calls from well meaning but extremely overprotective friends to let me know that “they can’t watch him anymore.” He is ten! The library, connected to the park, has a phone which he can use to reach us.
People called my parents hover parents, but at ten I could have played at the neighborhood park by myself.
I don’t think that CLIs are the path forward either, but you certainly don’t have to teach a model how to use them. We’ve made internal CLIs that adhere to no best practices and expose limited docs. Models since 4o have used them with no issue.
The amount of terminal bench data is just much higher and more predictable in rl environments. Getting a non thinking model to use an MCP server, even hosted products, is an exercise in frustration compared to exposing a cli.
A lot of our work is over voice, and I’ve found zero MCPs that I haven’t immediately wanted to wrap in a tool. I’ve actually had zero MCPs perform at all (most recently last week with a dwh MCP and opus 4.6, where even the easiest queries did not work at all).
The continuous exploits of MCP despite limited adoption really makes this seem wrong.
The story for MCP just makes no sense, especially in an enterprise.
To most people “I want to deport minorities” would imply nothing about citizenship status.
Someone with the opposite opinion would frame it as “open borders”, which is an extremist viewpoint globally and also not what people on the left in the US are advocating for.
Media coverage in the US is partisan. This is not an insightful viewpoint or nearly as incendiary as you’re making it out to be.
I guess the games you think are stupid depend immensely on your priors.
“It’s so dangerous, we’ve reached AGI, we just have to release models that are obviously incapable of abstraction for your safety”
In nemotron the high perplexity solutions are selected for RL, in VLM training a few people are looking at the entropy distributions of the training set, etc
You should probably argue your actual position instead of “your guy bad my guy good”. This comment is more Reddit than HN.
People play back the “Greatest Hits” without really going into the historical misses. The reality is that the quality and predictive power of science is covariate with culture.
There are a lot of good reasons to think that academic culture right now has a groupthink problem, mostly because the group is so much larger. Alternative theories typically have to wait for an incumbent group of thinkers to die. But if the gradient of thought is more continuous then do bad ideas become more sticky?