siliconc0w

I want some useful memory but it seems hardcoded to try shoehorn in personal details or tidbits from past conversations into responses. Even if I specifically ask it not to in my personalization prompt.

I've churned off OAI in favor of maxing out Claude. Better coding model and less creepy engagement hacking in the chat interface.

Used the "lite" models like Gemini flash - I hope if we do hand over the controls to the nukes we splurge for the top tier thinking model.

In other news, the lack of demand makes them pretty cheap at the moment. You can find a 2023 HW4 Plaid Model S for <$50k.

If Corporates can offshore their IP I should be able to offshore my likeness and rent it back to myself to reduce my personal taxes.

This was the old Google, new Google lays off teams and moves them to India. New grad hiring is all but paused, hiring just enough to Seniors to backfill attrition and trying to slot them in at n-1 or n-2 levels.

Google has a hugely valuable dataset of changes from decades of changes from top tier software engineers but it's so proprietary they can't use it to train their external models.

IMO AI is going to make buying things you need for your household without going through the clownshow that is Amazon a lot easier.

Even with the latest SOTA models - I still consistently find issues. Performance, security, memory leaks, bad assumptions/instruction following, and even levels of laziness/gaslighting/dishonesty. I spend less time authoring changes but a lot more time reviewing and validating changes. And that is using the best models (Opus 4.6/Codex 5.3), the OSS/flash models are still quite unreliable at solving problems.

Token costs are also non-trivial. Claude can exhaust a $20/month session limit with one difficult problem (didn't even write code, just planned). Each engineer needs at least the $200/mo plan - I have multiple plans from multiple providers.

* Tariffs are hurting most real-world jobs - you don't know what your inputs will cost in 6-12 months and so can't plan

* Political Corruption - you don't know if your permit, M&A, or regulatory rule-change will go through without a bribe to the right person

* Tech, which makes up most of the market, is all in on AI - these companies are late stage and the AI narrative is the only growth story so the market doesn't reward them for investing in anything else

* Cheap globally accessible labor

* Lack of enforcement of anti-trust means stodgy uncompetitive markets with players that make their margin through rent-seeking rather than increasing production or quality of goods and services.

* Poor investments and corruption in our healthcare system make it really expensive to hire Americans

* Poor investments and corruption in education make many Americans unsuitable for high skilled work.

It's not hard, just say you're looking for competitive compensation for someone with your experience and expertise and anchor on the high end of the range.

That said, what you say isn't as important as what you do. If you're interviewing, you should be interviewing at at least a few places to get comps. You will get lowballed if you don't have comps. Once you have a few offers, do one round of playing them off each other, decide which company you want to work for, and then say you'll sign today if they can beat the current best offer.

Most SaaS companies eventually veer towards the heavy enterprise which results in them raising prices and excessively complicating their app. So there is opportunity for a competitor but I'm not sure OpenAI is in a good position and it'd cement the view that they will eventually use your data to compete with you.

If you hire good people and set proper incentives they will figure out the best way to do something. If leadership has to direct employees to use AI it's a bad sign. If it were a huge boon to productivity, you won't need to force people to use it.

This is a good idea but I feel like you could get something similar by just adding an instruction for the agent to summarize the context for the commit into a .context/commit/<sha> file as a git hook.

If we actually taxed fossil fuel producers what it took to offset the negative externalities offloaded to the public, we'd be 100% on renewables long ago.

Sad to see we're going with a "child by default" internet. It'd be so easy for device and OS makers to align on an API that could tell the browser/app whether the user is under 18 or not.

same (at least for now, Codex seems to be much more token efficient)

You can also buy the hardware and hire an IT vendor to rack and help manage it as smart hands so you never need to visit the datacenter. With modern beefy hardware, even large web services only need a few racks so most orgs don't even to manage a large footprint.

Sure you have to schedule your own hardware repairs or updates but it also means you don't need to wrangle with the ridiculous cost-engineering, reserved instances, cloud product support issues or API deprecations, proprietary configuration languages, etc.

Bare metal is better for a lot of non-cost reasons too, as the article notes it's just easier/better to reason about the lower level primitives and you get more reliable and repeatable performance.

Wisdom is a thing, the longer you spend in tech the more you realize that most engineering work is probably a net negative.

I think opensource is a good analogue here. For many SaaS products, you don't even need to vibecode anything - there is already a reasonable OSS alternative. Yet people still pay for the SaaS. They want support, maintainability, security, edemifcation, a throat to choke, regulation and domain expertise, etc.

I do think like this HN post (https://news.ycombinator.com/item?id=46847690) is a good example of where a custom more domain specific solution makes a lot more sense that dropping in an off-the-shelf ERP. Still though, I think the bakery would prefer to buy the bakery-ERP than build it but vibecoding does reduce the barrier to entry so we might see more competition and share taking from incumbents by domain-specialized new entrants.

It doesn't make sense to 'build trust' with a bot. Today it works but tomorrow someone may push a malicious 'skill', a dependency may be compromised, or someone eventually figures out the right prompt injection incantation to remotely drain your accounts.

I don't see the demand for space being there, OSS is driving costs down and there are still plenty of hardware and algorithmic optimizations we haven't deployed yet.

The law doesn't really matter to this supreme court. They've already struck down most of Bivens and if this gets pushed they'll kill it for good.

American cloud companies will sell you a sovereign cloud solution but these are still pretty much make you a vassal state

It's pretty insidious to think that these AI labs want you become so dependent on them so that once the VC-gravy-train stops they can hike the token price 10x and you'll still pay because you have no other choice.

(thankfully market dynamics and OSS alternatives will probably stop this but it's not a guarantee, you need like at least six viable firms before you usually see competitive behavior)

Even outside skills, prompt-injection is still unsolvable and the agents need credentials to do anything useful so these things are basically impossible to secure.

The new shit-coin-as-a-service app(Bags) is a fascinating evolution of the system. Shitcoins started as a mechanism to monetize your own fame but have apparently evolved so you can monetize other people's fame.

On one hand this is pretty obviously dumb but on the other maybe I'm just not 'getting it' and if shit-coin-speculators want to help finance OSS projects (vibe coded or no) why complain about it?

Good for them to design and publish this - I doubt you'd see anything like this from the other labs.

The loss of competency seems pretty obvious but it's good to have data. What is also interesting to me is that the AI assisted group accomplished the task a bit faster but it wasn't statistically significant. Which seems to align with other findings that AI can make you 'feel' like you're working faster but that perception isn't always matched by the reality. So you're trading learning and eroding competency for a productivity boost which isn't always there.

I think something like an escrowed fee that both the applicant and the employer pay would be a reasonable way to solve the spam and keep both parties honest. If either the applicant or the employer are unhappy with the process (resume doesn't match, employer ghosts) - the fee is sent to charity, otherwise the fee is returned to both parties.

We're basically in a low-key recession that is being masked by circular AI deals and speculation.