mattbrewsbytes

There is a huge difference between Anthropic and Google dog-fooding: Anthropic does not own or build data centers, they rent (as far as all the reporting on this I've read). Anthropic employees dog-fooding with Claude tools isn't "internal", its most likely being run on external AI data center compute, which they have to pay for, there's just no margin on it. It might even be run out of Google's data centers.

I don't think you're understanding or reading what I linked to. The entire innerHTML of the page is not getting updated. If it was a page showing emails and there were next/prev buttons, next could trigger just the div (or area of the page that shows the list of emails) to reload with the server determined next 25 emails in an HTML snippet. The navigation areas don't change at all on the page, the page and browser don't go through a full refresh at all. There isn't a payload of JSON objects to parse and store in memory, nor a set of APIs to build on the server that output JSON.

Frontends have not exploded in complexity, they are still manipulating a DOM like they have been for 25 years. Complexity was writing different JS implementations for different browsers like Netscape Navigator 4 and Internet Explorer 4. I think a lot of frontend complexity is simply because of JS framework proliferation and a stance that the entire application should be JS in the browser, a full MVC framework when the backend is likely a full MVC framework already and has to validate all the inbound data changes.

What we're arguing about is mostly preference at this point. Please check out what different server frameworks do, there are server side JS frameworks too. If you're building simple web sites (like a pizza restaurant) vs. building a web application (like an app that processes millions of financial transactions per day) then the approach is going to be different and there are trade-offs to be made when picking a tech stack.

There is JS in the browser and will update a small part of the page by requesting HTML snippets from the server, not JSON payloads. Look at the Turbo+Stimulus framework. Its shipped as a library in Rails but it can be standalone - I built a small local webapp with it and the backend was Golang.

Read more here: https://turbo.hotwired.dev/handbook/introduction

Elixir has a similar mechanism with Phoenix LiveView.

I'd caution against "most diverting and pleasurable", thats really subjective.

For me woodworking or similar DIY projects making things with my hands (i.e. not 3d printers - not that there's anything wrong with that but most of the creation there I think is on a computer). I took architecture drawing classes in high school and I like the pencil and paper element of that for projects. With wood, its very malleable so if you make mistakes you can really easily adapt and keep going.

The unexpected benefit I found is when you make things for other people, they really appreciate it and tell you as much. They almost think its magical like when you fix their computer but they are much more appreciative because there's a physical thing there they use, maybe daily, maybe others comment on it and ask where they got it.

There are server side frameworks that use web sockets so you can refresh small parts of a page and not the entire page. This involves smaller round trips from client browser to web server. In my opinion, for data intensive applications it makes more sense rather than REST APIs returning JSON payloads, parsing JSON into objects client side, and then update the DOM.

I think a lot of JS developers fail to recognize the variations of client systems power or people's usage patterns like how many tabs are open, how much GB of memory is their browser using, is the client machine paging out to disk, stuff like that. Not everyone is running a MacBook Pro so heavy client side JS apps tend to be "slow" to end users, at least perception.

There are some similarities between Anthropic (or OpenAI) and WeWork. Obviously the large difference is software but there is a whole lot of leasing hardware going on (neither frontier lab owns any infrastructure as far as I can tell), long term capital commitments, short term monthly subscriptions, no real moat for the frontier labs (or a hip office rental space), etc.

I think 2027 will be when a lot of companies revisit their AI spend and ROI. Have there been measurable business benefits reported by the media outside of programming?

Is there an accepted definition on what a harness is? Is this like GasTown and similar? Plan mode? Is this a client-side loop with multi-threaded agents running in parallel? If the AI client is doing this on its own, is that considered agentic and/or using a harness? Or does the introduction of saving text to a MD file make it "self learning" a harness?

It feels like the pace at which the industry is defining things may be faster than the actual pace of change sometimes.

The laws changed for IPOs. It used to be when the total investors got over 500, including employees holding shares, that would trigger a company to need to go public. It was changed some time ago (not sure if JOBS act or what) to be much higher, like 2000 investors and excludes employees.

This means companies stay private and keep getting private capital/investments (the flip side being private equity in the pre-IPO company) and typically an IPO is really existing shareholders realizing their investment.

So Amazon, Microsoft were forced to go public much earlier, then had their growth. Anthropic has already had growth, Claude AI is well known at this point, and after IPO there may not be as much growth after. In fact, the stock might go down because they have to report their full financials.

I think another element the cost of maintaining it, supporting it, marketing, etc. hasn't been commoditized like producing brand new software. Those things take time to do properly and there are lots of people churning out a SaaS and then they get abandoned because its not fun anymore to do all the administrative/sales stuff.

I seriously don't think non-technical people are pumping out apps, probably never will. They simply don't have an interest and would rather just pay for a service. A HVAC, plumbing, or electrician company isn't going to write their own CRM system. Now they will price hunt those services so as software production becomes cheaper, I would expect SaaS margins to also be going down.

Selling tokens is a similar approach to social media "engagement" loops of keeping people in their app. It works but the detractor for frontier labs is no ad revenue.

I think you should read up on Xerox and PARC. A single company can only take on so many things at once and be successful. Also it was ~60 years ago so at that time they could only manage the IP of laser printers as it aligned with the copier business. They didn't productize GUI, Ethernet, and other inventions or concepts like tablet computing that came out of PARC.

Now there are conglomerates like Alphabet that have Waymo, YouTube, etc. very different businesses, thats a bit different. It might actually make business sense for the frontier labs to do something like a CRM system as a spun-off separate company. That might be in direct competition with their source of capital and probably would violate some agreement - if they have a non-compete type of setup with the likes of Oracle, Google, Microsoft.

People running old versions and not applying patches rigorously is also normal. Its been normal for decades, nobody likes doing maintenance so it gets forgotten or ignored.

For like 95% of web applications, they probably could be ported to nearly any other web application tech stack with the tools we have now. We've been using HTTP GET|POST|PUT|DELETE for almost 30 years with HTML, CSS and JS on the frontend (with interludes of Flash, Java, etc. in clients). I think this AI hype cycle is going to lead to the tech industry spinning their wheels for a few years rewriting a bunch of things "just because I can" much like the industry spawned countless Javascript frameworks for a while (surprised we haven't seen more yet).

Instead of discussions about "my language/framework is better than your language/framework, see my TrustMeBro benchmark" - why not figure out what's next? Could we be working on the next set of patterns that are a level higher than OO design patterns like MVC, AR, etc.?

You could say the same about any language or framework.

I wonder how all the old PHP apps are going to fare today, I hear there's a major security issue.

It feels like another layer of "conceptual compression". I'm not sure who/when that phrase came around but in context of Rails is where I heard it and its things like ActiveRecord or really any ORM. You aren't writing the SQL directly, something else is and for the most part it works. Then it doesn't work or your database grows too much and you need to optimize table structures with indexes, etc.

My point is if you don't understand the fundamentals of the underlying code (SQL in the example above), you're more likely to have issues in the future. Or maybe not if that app doesn't reach a usage pattern or complexity where the nuances under the hood matter.

It really depends on what you're shipping. Meta ships social media apps where in the grand scheme of things if something is broken software wise it largely doesn't matter. What's the worst case? lost ad revenue? Users can't do things? If that were software for MRI machines or aerospace where software issues/bugs can result in lives lost then it matters much more in understanding what is going on at all levels of the software and potentially hardware.

If LLMs were brains, it would be someone with severe amnesia who you had to tell them what an arm and a leg were every time. Its stored vectors of words or partial words and computations of math that run to predict the next words/partials to use to form text output. It will nearly always find the real sets of words that are load bearing to put into its output (seriously how many times do real and load bearing appear in the models, can we remove some of that).

What's interesting is most of the progression we've seen this last year in AI agent clients are very basic computer science concepts: loops, threads, arrays (of JSON) and memory management (aka context) with varying combinations of those put together and people in computing are writing about it like its some huge break through. I would love to hear a game developers perspective on comparing AI clients (not the server side LLMs/vectors/maths) with games - they seem to have very similar computing problems.

I think its easier to state "bullshit jobs" but the reality is every single job has "bullshit tasks" and usually these are the easiest things to do. TPS reports and the like. BS Tasks are ripe for AI to do the drudgery but I don't think there are a lot of actual full time jobs where the entire job is BS. These can't exist en masse for very long in capitalist companies.

There's always Dunbar's number - the larger an organization gets beyond the capacity of people to socially know one another, the greater the "what do they even do?" types of attitudes form about other distant parts of an organization.

An interesting difference between the AI boom and the dot com boom is that back then companies would IPO to raise capital and might have been forced to IPO when they were beyond 500 investors incl employees. These days these companies have already raised capital via private equity and the laws changed which allow them to not count employees with equity and have much higher investor counts. An IPO by either of the frontier labs companies is really a cash-out and payback on the capital they have borrowed. My thinking is the stock price would go down, possibly significantly, after an IPO.

Oracle's credit rating is hovering just above junk bond level. I think the capital agreements are similar to an Adjustable Rate Mortgage and monthly so if there's another rate increase from the Fed and Oracle can't find anyone to buy their bonds and/or their free cash flow continues to be negative then maybe that is a pin-prick in the stock bubble? Its sort of like if you keep bouncing around from 0% credit card offer for 12 months to another and then you stop getting 0% offers, then its time to face the music (those usually have a 3% transaction fee, at least a long time ago they did).

I'm not a zero computers zen wood shop person. Just offering a perspective based on my experience. I've also built from others digital plans and my own and find that for my workflow, the higher fidelity doesn't add much value. Maybe its a personal tendency to try and make it pixel perfect and because of that I sink way more time into it than I should. Pencil and paper help me avoid that rabbit hole for woodworking.

You really need to build things to learn and progress. Woodworking for Mere Mortals youtube channel is a good place to start. Start with hand tools and build some very basic things, see what you like doing and progress from there. The caution with all online content is the constant promotions, etc. for recommendations of tools or brands. Look for counter opinions. See if you can go use a tool somewhere before buying, maybe at a local store like Rockler or Woodcraft - they may have classes. What you won't see in online content is how long something really takes, they are probably editing 40 hours of video down into 30 minutes.

In my experience, a track saw is immensely more useful than a simple circular saw or miter saw for finer woodworking or cabinet work. The other two are great for rough carpentry but tend to be harder to use for finer tolerances needed for finer woodworking. They are more expensive but because of the track you will get straight cuts with little setup. I've had issues with a contractor sized table saw even with it built into and secured in a DIY cabinet. Lifting and moving large sheets of plywood or assembled boards like a table top can be prone to more errors than setting up and moving a saw along a track. Really expensive table saws are probably better but as a hobbyist I don't have $3000 for a single tool.

Why ruin the fun of a hands-on project with computers? I'm also a hobbyist woodworker/DIYer and enjoy the drawing aspect, thinking through the space, the dimensions, etc. and drawing that out with pencil and paper. Its relaxing and by nature disconnected. I'm not on the clock and trying to build things for customers so speed isn't necessary. The level of accuracy in software is thrown out the window when you start dealing with actual wood that has grain, knots, different sizes, etc. I find myself making adjustments in the shop as a project progresses where fine measurements on paper are mostly useless.

I would argue if you're building plans to make a lot of repeatable items, then you're beyond hobby level.

I struggled with upper level math, anything beyond calculus. Can someone explain what technology is being used for things like this - doing mathematical proofs? Is it LLMs? Some other GPU thing? Quantum?

I have a hard time thinking a neural net "predict the next token" is going to do things like solve math proofs or find cures for cancer when it gets very basic things blatantly incorrect sometimes.

Oh totally. This is a playbook for a Fortune X00 company. Layoff your staff and create a "services" SLA contract with a third party company to deliver IT services. Just the legal team to get thru all the docs alone means you need a legal and procurement/contracts department. Not for startups.

1. Risk - you throw an SLA on services and can point fingers when shit hits the fan

2. Circular agreements - you get them to sell your goods/services in the region of their home country

3. These count as "Services" on the P&L and provide massive tax advantages to the company because these aren't treated as labor costs.

This may be from an old context in my brain since I worked with some of those companies, so those reasons may have changed or impact may have with various regulations, etc.

One could describe similar issues with video/youtube content. Everyone is engineering it for the algorithm but the thing humans want to know up front should be in the first 30 seconds.

I plan to use english, but not 1500's english.

That might be a bad example ... replace english with nearly anything not related to prompting/coding. For example, I bet the models have "knowledge" of biology, chemistry, etc. not exactly useful for programming a SaaS web app that say does project mgmt. I think there's opportunity for very specific tooling rather than "general" knowledge.

I still have this naive notion that we don't need LLMs for code generation and editing. Does a system need the knowledge of the full works of Shakespeare to be able to output Javascript?

Maybe people smarter than me know better but couldn't there be a middle ground where an IDE/Editor has an embedded engine (doesn't need to be a full-on LLM) that doesn't require external tool calls and token spend?

If an organization is paying $2400/year per developer for tokens and a highly intelligent editor/IDE comes around that charges $1000/yr and gets more output at a fixed cost, its a no-brainer of a decision.

There was a run on Mac Mini's for people running OpenClaw earlier this year. I was doing some research earlier today on an idea and that came up as an option because it (and Apple's hardware for Macs) have shared RAM across CPU and GPU so its apparently a viable option to run ollama (ironically from Meta) or other local models for simple tasks.

I think VPS are going to be expensive because of the RAM and GPU requirements - 32GB RAM is beefy for a VPS and might run like $30/month on the cheap end. Thats $360/year so running something at home might be more secure, capable and cheaper. Remote/mobile access might be viable with something like tailscale.

I think some of those tasks are part of the fun of a trip - doing the research on locations, events, restaurants, etc. You stumble through a bunch of things and little nuggets of that sit in your brain and then when you're on the trip they might come up - one place is abruptly closed, something looks like a bummer in person, etc. Without doing that research yourself then on the trip you have a list of things and aren't really knowledgable about alternatives.

Like I said, I'm not the target audience for a personal assistant. That does sound like a more ideal UX than what chat gippity and the others offer though. I would expect that with Meta - they really get the UX mostly right for people.

Meta's active user base doesn't care about privacy.

Most personal tasks take trivial amounts of time these days. Does a 5th grade teacher or an auto mechanic need a personal assistant? Do we need to be able to book a trip faster? It feels very gimmicky to me, but I realize I'm not their target audience.

Given that, I do think Meta will probably get the lions share of B2C AI and they have a proven track record of using personal data to serve ads and make a lot of profit. If they capture more of their user base's time, it keeps them engaged and away from a list of competitors. If Meta starts crushing it in this fashion I could see IPO's of OpenAI and Anthropic land like fart in church.

I think we're a few decades away, at least, from seeing a robot on something like a residential construction site. Sure they look good in a flat controlled environment, but humans have a hard time walking around a construction site. Then there is the dexterity issues that they have an human hands excel at. The startup cost of a robot is probably 4x the cost of an entry level tradesperson, at least.