JavaScript is dynamic and supports live reload which means iterations are far faster than would be in a compiled language--even for LLMs.
This is especially useful when you're trying to evaluate behaviors while changing state surgically.
7,717 karma · joined February 23, 2013
This is especially useful when you're trying to evaluate behaviors while changing state surgically.
They've done this in other industries like solar panels, chips, and EVs. This is no different.
Systems aren't a single addition. They are compounded operations with sprawling complexity. What happens when you can't reason through the system? What happens when you start asking for the wrong things? What happens when saying "fix it" on loop stops working?
In the old days, sales would close a client with the help of someone in technical sales and a solutions architect that would come up with a high level spec for a solution. That spec would be fleshed out post sales and then given to an engineer to develop.
Today the FDE often does both.
The difference between Palantir and pure consulting is that Palantir has a core software platform that they customize per client. The business model is more akin to IBM or SAP products.
What would the difference be?
> these are the firms that spend on multiple models and use the most advanced and productivity-enhancing products available (coding agents and APIs as opposed to simple chat subscriptions)
> who funded you is a better predictor of AI adoption than the sector you’re in
My bet is that these companies are either:
1. In the business of trying to automate other jobs. The number of startups in this space is stupefying, so I'm sure more established companies with VC cash want a slice.
2. Leveraging AI to accelerate growth as much as possible because that's their mandate.
What's scary is that hiring across the rest of adopters (most employment) has remained stagnant. So what happens during the next recession?
1. Compute costs collapsed since the advent of Cloud and yet hyperscalers still have fat margins.
2. Many open source office suites exist yet none compete with the ubiquity of gsuite or office. GitHub, Slack are similar examples.
3. Both Windows and macOS dominate the home desktop space despite free alternatives existing for a long time.
4. Many formerly open source infrastructure components like Redis and Elastic Search have Apache equivalents, but they still command healthy margins.
I understand the arguments for a margin collapse, but I don't see any historical analogues. It seems that enterprises will pay top dollar for service guarantees, integration, and someone they can sue.
It's nobody gets fired for buying IBM all over again.
They have specials every now and then.
Since the very beginning I've ran Claude from an isolated VM on yolo mode. This is just like giving an engineer their own laptop. Claude works on a feature up to a PR worthy point. I review the diff, just like I would with another engineer, and massage it to get it in the right shape and move on.
Inexperienced engineers make the same mistakes described I've even seen rm -rf albeit not from root! I would have lost my mind micromanaging someone with all permissions denied.
We all need cyborg cats to hunt cyborg mice. There's no other compromise unless you want a rat infestation.
> This is not a malicious feature, but it is a weird choice for a developer tool that asks for trust.
They already tell you they scan for malicious prompts, and they have no ZDR guarantees for consumers. Why do signatures like this matter at all?
I had this experience doing a port from Big Query to Postgres using Opus. I had unit tests to guarantee parity with the original code, and Opus insisted on building this bespoke query builder (e.g. `def _where(very_complicated_params)`) on top of sqlglot.
Even with the original code being straightforward and legible and repeated instructions to match, I had to fight with it to get close.
In the end, I ended up doing things the "old fashion way" where I copied chunks code into Claude proper and gave explicit instructions for each piece.
I clearly had externalized the requirements, and yet that wasn't sufficient. The only way to unit test further would be to use an AST to evaluate the output against metrics I couldn't even encode.
Any reasonable company would be pissed if a competitor, especially at Ali Baba's size, leveraged that company's R&D to compete. It is in this sense, a corporate attack.
If you want to roll your eyes at distillation concerns, you might need to excuse Anthropic for originally using pirated material to train their models.
NASA spent something like $300B in today's money on the Apollo program, and Artemis has exceeded $90B already.
I'm much more keen on never getting sick than prepping for Mars.
The drug dealer analogy has a darker side to it, however.
Once your dependent, they can drive up the price just because. It doesn't need to be for existential reasons.
Moreover rents for affordable housing haven't kept up with inflation while benefits have.
Arm chair speculation like what's in the article won't suffice. People need to be surveyed and interviewed to get to the bottom of this.
> Only about 5% of denied in-network claims were turned down because the care was deemed not medically necessary. The rest were administrative, for an excluded service, for a missing referral or prior authorization, or for a reason the insurer never specified.
I worked in health tech for a while, and I can tell you the muck around a lot with ICD/CPT codes to maximize billing along with other shenanigans. There was actually a project at an innovation center at a well-known medical center which leveraged ML to maximize the amount of codes they could bill for without being rejected. The same kind of thing is often done by physicians who want to juice insurance.
Be mad--very mad--at hospitals and drug cos. As providers, they present themselves as patient advocates, but they're responsible for the outrageous healthcare costs. The dollar amount paid out by US insurance companies is maybe 2x that of other OECD countries, but the healthcare we get back from providers is trash (and extortive) by comparison.
If you haven't figured out a good abstraction at 5-100 customers, God help you.