I think 'actual parallelism' is a vastly easier and more fruitful way to get better performance out of these kinds of systems, compared to pushing against single-threaded faster generation. Tool calling and responses are often embarrassingly parallel. Code generation tasks naturally have a dependency tree that can be unrolled into a fixed budget of parallelism. Tasks can be hierarchically decomposed into subtasks.
It's the same asynchronous stream pattern we're used to dealing with in regular software engineering. We have a fixed thread pool, lots of work that can be scheduled concurrently. Since these are streams, we can do the compute incrementally to reduce the time-to-first-byte/token/response.
Since so many tool calls are inherently asynchronous, and subagent task decomposition can be modelled as such, the IO streams can be oversubscribed, and incoming responses can be priority queued.
On the intelligence front, it's incredible how much better frontier models perform when you just interrupt them every so often and go 'is that the best you can do?', or re-iterate instructions, or repeat the overall goal. I find instruction following _so poor_, especially for 'presentation layer' aspects. Yet if I ask the model to rewrite its last response, it does so perfectly. Why can't the model do this 'internally' and save me having to say 'try again'!
Just because the 'model' is autoregressive doesn't mean the system as a whole needs to present a single stream of immutable text.
I don't buy the "any constraints cause lower performance via being out of distribution" idea. Sure if you ask the model to output 'reasoning' in JSON steps, that is a completely different 'channel' to its trained 'reasoning' output. For real tasks though, I think it's more about picking the _right_ context free grammar to enforce format correctness. You can enforce an in-distribution format and get the best of both worlds. I don't think the industry should settle so hard on JSON-for-everything.
I was working on a speculative decoding optimisation and its accompanying blog post. Explaining the more basic concepts filled so much of the post I decided to pull them out, forming this article.
I had a bit too much fun with the tokenisation diagrams / animations. The raw text is provided to an Astro component, which tokenises it, and forms the individual DOM elements of the tokens. I find it really hard to read 'tokenised' text, I figured some consistent colouring would help. The 'Probabilities' component is a trivial grid, but all the other components support 'word wrap'.
I ended up writing a 'responsive design aware graph colouring solver'.
Multiple screen widths, 'desktop' and 'mobile' are 'simulated', forming an adjacency graph of tokens that touch. Colours are then greedily allocated, then optimised per page over a few hundred iterations, swapping allocations to enforce minimum hue distance between touching tokens at those common screen sizes. The optimising value function prioritises even distribution of colours, because it looks nicer than maximal hue difference.
Originally I naively outputted the palette styles per component, but found the css post processing optimisers didn't handle that as well as I'd have thought. So then I wrote a little 'CSS compiler' that takes the high level palette and timing concepts of the animations, and optimally merges rule declarations.
The start of the post really relies on the animation occurring while fully in view, so I set up some IntersectionObservers that do the 'please scroll' text.
I tried my best to have it all work when JS is disabled on the client. I tried to get the 'hovering' to be CSS-only, but found the JS solution much more performant.
The DAG diagrams are formed with this neat Needleman-Wunsch algorithm from the bioinformatics field. The Astro component accepts several 'examples' then aligns common subsequences, producing the CSS grid and the 'basic SVG' on the server. The responsive nature meant I had to move the final 'allow' generation to the client.
Some browsers seem to throttle the token animations sometimes but I haven't figured out what causes that. This is my first time leaning hard on CSS variables.
In South Australia an algal bloom started in ~mid-March of this year, it's a pretty big ecological disaster, probably the worst non-bushfire disaster in living memory. Probably 30% of SA's coastline is affected. It's a pretty big deal affecting many people's livelihoods.
The joint state and federal government relief and cleanup package is worth AUD $102.5 million dollars.
I hope the public receives that comparison at every opportunity.
The old website was frankly excellent, the only problem was it didn't have HTTPS support. I would have happily upgraded that part of the system for the cost of a cup of coffee if I'd had an opportunity to submit for the tender!
The new website is significantly more difficult to navigate (for me, a seasoned tech user). The primary thing Dad's everywhere use it for (the weather radar) now requires scrolling to the _bottom_ of the page, and zooming in from the 'map of Australia' to the region you live in. It used to be like, a click to go from home page -> state weather radar with all the info you needed.
> [BOM] said the cost breakdown included $4.1 million for the redesign, $79.8 million for the website build, and the site's launch and security testing cost $12.6 million.
Absolutely stupid, even those numbers are outrageous. They say it's part of some 'larger upgrade package', prompted by a cyber attack in 2015.
But politicians over here love to blame cyber attacks when technical blunders happen. We had a census a couple years ago and the website fell over due to 'unprecedented load' or maybe it was a 'DDOS attack'? The news at the time couldn't decide who to blame!
Welp, I hope this gets as much world-wide attention as possible so they can be embarrassed and do better.
It would be nice if users of the codex-cli that are just using API keys as a way to handle rate limits and billing could receive these new models at the same time. I appreciate the reasoning behind delayed 'actual API' release, but I've found the rate limiting to be quite annoying, and my own API keys don't have this limitation.
Apart from the actual model release, this is the second set of models from OpenAI that uses the Harmony response format. I don't suppose anyone knows if OpenAI uses the Harmony format internally for GPT-5 as well?
Not at all, in a renderer the Node and Chromium event loops are bound together, they’re part of the same v8 isolate, no IPC shenanigans.
The main process really shouldn’t be used for anything except setup. Since it controls gpu paints amongst other things, blocking on it will cause visible stuttering and a bad user experience.
Separately the IPC lets you do zero copy in some circumstances via Transferable objects such as ArrayBuffers. Structured cloning is efficient but not zero copy, and json serialisation shouldn’t be used (since structured cloning is easily available).
You don’t need IPC, you can either use a preload script to expose particular Node APIs in a secure manner or set ‘nodeIntegration‘ to ‘true’ to expose everything.
Within a renderer you can access NodeJS APIs directly. The main process shouldn’t be used for any significant computation, as it will block GPU paints and cross-process synchronisation.
The other main difference is Electron bundles a known set of APIs, given the known Chromium version. There’s such a huge variance of supported features across the embedded web views.
Conversely, the last blog post we wrote was 8,000+ words and took months of testing, yet the average 'read' time is under 2 minutes. I'm convinced there's a correlation between interested technical users and the blocking of analytics scripts - but if I were to naively look at the data, I'd also come to the conclusion that "lower effort" was better return on investment. I wonder if these tech journalism establishments are following their analytics and A/B testing themselves into oblivion.
We nerd sniped ourselves into testing the latencies of a whole bunch of wireless communication links and protocols for microcontrollers.
We’ll probably do a series of power consumption / range tests later on, let me know if there are any setups in particular that you’d be interested in seeing test cases for.
Raw data, firmware and post processing scripts are here on GitHub:
I work on a product for building user interfaces for hardware devices. All the state management is done via incrementally updated, differential DataFlow systems. The interface is defined in code instead of graphically, but I think that's a feature, so that code can be version controlled.
I think there has been evolution in the underlying data computation side of things, but there are still unsolved questions about 'visibility' of graphical node based approaches. A node based editor is easy to write with, hard to read with.
Our product, Electric UI, is a series of tools for building user interfaces for hardware devices on desktop. It has a DataFlow streaming computation engine for data processing which leans heavily on TypeScript's generics. It's pretty awesome to be able to have examples in our docs that correctly show the types as they flow through the system. I certainly learn tools faster when they have good autocomplete in the IDE. Twoslash helps bring part of that experience earlier in the development process, right to when you're looking at documentation.
Our site is built with GatsbyJS, the docs are a series of MDX files rendered statically, then served via Cloudflare Pages. We use the remark plugins to statically render the syntax highlighting and hover tag information, then some client-side React to display the right tooltips on hover.
We build a Twoslash environment from a tagged commit of our built TypeScript definitions, from the perspective of our default template. The Twoslash snippets as a result have all the required context built in, given they are actual compiled pieces of code. The imports we display in the docs are the actual imports used when compiling from the perspective of a user. It bothers me when docs only give you some snippet of a deeply nested structure, and you don't know where to put it. Even worse when it's untyped JS! Using Twoslash lets us avoid that kind of thing systematically.
The CI system throws errors when our docs snippets "don't compile", which is very helpful in keeping our docs up to date with our product. Nothing worse than incorrect docs!
We use React components extensively, and I'm not really happy with our prop reference tables which use Palintir's Documentalist. Our components are increasingly using complex TypeScript generics to represent behaviour. The benefits in the IDE are huge, but the relatively dumb display of type information in the prop table leaves something to be desired. I'm most likely going to replace the data for those tables with compile-time generated Twoslash queries.
My only complaints have been around absolute speed of compilation, but I haven't dug deep into improving that. I just set up a per snippet caching layer, and once files are cached, individual changes are refreshed quickly. After all, it's invoking the full TypeScript compiler, and that's its biggest feature.
Overall I've been very happy with Twoslash, and I'm looking forward to refactoring to use this successor and Shikiji (the ESM successor to Shiki), hopefully it improves our performance. The new Twoslash docs are look great, a huge improvement on when we started using it.
It's the same asynchronous stream pattern we're used to dealing with in regular software engineering. We have a fixed thread pool, lots of work that can be scheduled concurrently. Since these are streams, we can do the compute incrementally to reduce the time-to-first-byte/token/response.
Since so many tool calls are inherently asynchronous, and subagent task decomposition can be modelled as such, the IO streams can be oversubscribed, and incoming responses can be priority queued.
On the intelligence front, it's incredible how much better frontier models perform when you just interrupt them every so often and go 'is that the best you can do?', or re-iterate instructions, or repeat the overall goal. I find instruction following _so poor_, especially for 'presentation layer' aspects. Yet if I ask the model to rewrite its last response, it does so perfectly. Why can't the model do this 'internally' and save me having to say 'try again'!
Just because the 'model' is autoregressive doesn't mean the system as a whole needs to present a single stream of immutable text.