starfallg

Good idea! Let me roll that out. I guess you can also print to PDF.

Thanks! I started on this last week in my spare time and I used Claude, all my commit history is on Github. I did most of the coding via Claude Code, no fancy system or exotic harness.

I would love to do more by hand, but I look after a tech team inside a renewables firm, so I don't have much time between job and my preschooler.

You can play around in the guestbook. I'm in it right now and you can see me editing things.

Cloudflare Durable Objects are pretty cheap and I'm using it like a very basic pub/sub with each deck having it's own DO. I did some calculations and it should be well within my spend limits.

Yeah, if you do a lot of slides, this is for you! The whole point of this is to make doing slides bearable!

Thanks for sharing this! Let me take a look. I was only aware of TiddlyWiki when I first started on Bento.

Thanks! I guess there are some drawback to this approach, large companies especially would be completely adverse to it. But it's cool and I'm hoping people would find it useful!

It's really to solve a use case where there are a lot of these cool web-based presentations being created by AI, but they are all disposable things that non-technies can't work with easily.

Base64 is paired with compression to keep the size small so you can share it easily. The original version was around 1.3MB and maybe we can have a switch to output without compression. The source that generates it is all in Github.

Hi, I'm the creator of Bento. Just wanted to share a bit more about how I created it beyond what's in Github.

The file contains more or less two sections. There is a plain block of JSON near the top of the file which is the slide data. You can read, grep, or point a harness at it. The app itself is in a base64 blob that loads through a small shim which deflates in the browser with DecompressionStream, which keeps the package small and so that we don't need to fetch any external files at runtime. File System Access API is used for JSON writeback into the same file, which falls back to a plian download and all updates are ECDSA signed.

I started with reveal.js which served as a base, and incorporated GSAP/Flip to handle the animation. Then I added charting with echarts, but due to a number of issues with size and how the worked, I ended up re-implementing both.

What I'm most pleased about is how seamless the CRDT works. The blind relay is a small file that runs on Cloudflare Durable Objects, and all it sees are the encrypted data from the clients. Collab is off until we turn it on in the app (by starting a session by sharing by invite file), and then some sets of keys are generated. Access is handled by user key, and you can have read-only users, and also revoke live collab access by user as well.

You forgot smartphones, where low-cost did not win out. It led to low margins for the Chinese firms and eventually left them unable to invest properly in key markets. They may still hold marketshare, but in terms of profits, falls well short of Apple and Samsung.

I can see a lot of parallels here. Model performance doesn't matter if you can't make the system commercially sustainable.

More to do with how the ARM ecosystem works and the resulting lack of openness and standardisation in the hardware interface.

CDBC doesn't necessarily imply blockchain. It can be just another payments system.

Chatbots are good at helping humans arrive at the specification. Apart from people that has deep expertise and/or talent in this area, humans aren't very good at building these specifications. That's the topic of this exchange.

I guess the flip side to that is the iterative conversation to develop these specifications is exactly what AI chatbots are good at.

It's also the main reason why very structured AI agent orchestration for software engineering modelled on rigid processes fails to really provide much value.

Nope, just break up the one that has been consistently found to be abusing their market position. Microsoft has been embroiled in this since the 90s.

Exactly. You might not like what Google does, but you can't deny it's a massive commercial success. Just because their approach to creating and delivering apps might not be to your liking, you might actually be the niche.

Yes, zero Forex fees cards work. But the terminal detects that your home currency is different to the local currency and you still have to choose the right option.

For example, just the other day I fat fingered the screen and chose the wrong currency.

That's the nature of decentralised control. It's not just DNS, phone numbers work in the same way.

Combination of factors, but mostly the success Douyin had in Mainland China leading to the investment in TikTok internationally, given that no other Chinese social app had reach this level of penetration.

Vine had more active users than Musically, and look what happened to Vine. It would have never had the investment internationally if not for the blow-out success it had inside Mainland China.

TikTok is a fluke, created by the condition of how it was originally born as Douyin in China. It is also the only app that translated well from the domestic China market to international markets.

Being backed by lots of VC cash and Bytedance's revenues in China is a key factor in getting TikTok established overseas.

The reality is that hacking code isn't always beautiful. Most of the time, it is mundane grunt work.

You can always leave the core logic for your to work on and have the AI handle all the bits that you don't like to do. This is what we do for modelling for example, AI helps with the interface and data backends, the core modelling logic is hand-crafted.

Pretty sure we can make LLM agents to transform declarative inputs to agentic action.

The rate limits are not because of compute performance or the lack of. It's to stop people from training their own models on the very cutting edge.

Translation between languages and into logical representations is one of the main goals of NLP since the 1950s.

What makes you think that Google hasn't developed natural language processing, when they launched Google Translate in 2006?

Also, what makes you think that NLP would solve the problem you're describing? Only LLMs has proven that it could fully understand and process the queries in the way you're describing, hence why I brought that up. However it is expensive computationally, and only recently was it even technically possible to do.

>Easy. Google is basically spyware. It's an advertising company, and their product is you.

This is such a flippant and facile response.

Google isn't a advertising company, it is a tech company that gets the majority of its revenue via selling advertising. This can change, and also likely to change in the next decade or so.

The reason things are is that nobody was willing to pay for search - it's a product with a very low incremental cost. The market dictated this operating model, and nobody has been able to upend this model so far, not even OAI. The numbers just don't work out. Do you think OAI can continue to subsidise free ChatGPT queries with paid ChatGPT Plus subscriptions? Almost certainly not.

> Because one of the most common things I search for is stuff like 'what is the function in [x] that does [y] in [z]'.

This is also the type of search that Google makes no money from.

The money is in searching for up-to-date relevant product information, where Google is the undisputed leader.

>Search engines, especially with the resources of Google, could have developed at least basically functional natural language search decades ago

Google is one of the major AI research outfits, and arguably the only one that continues to deliver consistently over the last 2 decades. Statistical Machine Translation/Google Translate, Adwords Quality Score, TensorFlow, AlphaGo, Attention is all you need (Transformers), AlphaFold all Google innovations.

You can't really blame the prevalence of SEO slop on Google. It's not the lack of want of trying, it is hard technically (see how long it took to develop modern AI capabilities), expensive computationally (as we can see with the unsustainable cost of test-time search in ChatGPT) and in terms of user-experience.

>Google is certainly well on their way to becoming another Yahoo

Really, it isn't. Google is in the unique position of being the closest technology company to achieving full vertical integration of their value chain, from silicon to software to data to end-users. They are also at the forefront of frontier AI, including productising the research output. I don't really get the Google hate on HN, apart from maybe the YC/sama bias.

Bingo. Because it's just another Claude Code fanpost.

I mean I like Claude Code too, but there is enough room for more than one CLI agentic coding framework (not Codex though, cuz that sucks j/k).

>The individual writing this guide appears to have only had barely two years in a hard 'management' role, according to their LinkedIn. How can they possibly be qualified to give advice on a subject that takes years and years and years of developing and refining soft skills, let alone consulting on the particulars of leading people in that time? Might as well run for President. What arrogance thinking they can go into an organization and infect it with the Tech-leadership-style-du-jour and think they've done some good.

The skills in leadership don't start developing when you become a 'manager' at work, it starts developing around the time you lose your baby teeth, maybe even before. From personal experience, there is little correlation between the amount of time spent in 'management' and how much one understands leadership.

> In practice the best managers care primarily about politics and growing their personal stake in the organization.

This is fine, and expected of senior management.

> I've found that the more clueless they are the better (just don't point that out).

This is toxic. Management at the senior level of these organisations has been completely detached from the activities on the ground.

A healthy functional organisation should have senior managers that understand the complete stack of layers, but especially at the level where the value is being generated they need to fully understand.