On my M1 MacBook Air with MacOS 12 it hangs. The mouse pointer still moves, but I can't bring up the force quit dialogue, and couldn't get anything else to work either. A youtube video in another tab kept playing until I held the power button to force a shutdown.
Github is a terrible platform. We have had multiple enterprise tickets open with no response for multiple weeks after they fraudulently charged our company for services we cancelled weeks in advance. If you value any kind of real customer support or response time, I would avoid Github like the plague. They will steal your money and there is no one to even talk to about it. Their customer support is non existent (probably because their platform is in shambles).
Github is a terrible platform. We have had multiple enterprise tickets open with no response for multiple weeks after they fraudulently charged our company for services we cancelled weeks in advance. If you value any kind of real customer support or response time, I would avoid Github like the plague. They will steal your money and there is no one to even talk to about it. Their customer support is non existent (probably because their platforms is in shambles).
Fantastic idea! I think the hardest part for people to adopt this will be overcoming atrophied handwriting skills and no autocorrect.
One thing that doesn't seem clear to me from reading without signing up is how do you actually upload content? Do you have to have paper and pen and then take a picture of it, or do you provide some kind of excalidraw-like canvas for writing? I notice there isn't any paper background or line rules, which I think would also be nice for those who fancy a bit of structure when reading/writing.
Unfortunately I really don’t understand enough to judge about that. Could you tell me some things that made you go
with a “buzzword salad” sentiment? Not as a gotcha or anything, just out of genuine curiosity.
The FPGA portion (second bullet point in the abstract) reads like research à la Taalas [0] (known for ChatJimmy [1]), so it’s not immediately clear to me why this would be indicative of a lack of knowledge of the field.
Here are some easy to ways to tell if you are reading AI slop. Any one of them should cause a flag, but all of them at once should sound the alarm.
1. Investigate the author. In this case, it is an administrative scrivener & independent researcher. Nothing wrong with that, but credibility just isn't there for such dense CS, math & "physics".
2. Check for peer review. In this case, there is none. Published on zenodo, a platform notorious for self-publishing authors plagued by runaway AI sycophantic psychosis.
3. Check the author's publishing cadence. In this case, we are already at version 8 in less than a week. In traditional research, iterating from v1 to v8 represents weeks/months of peer feedback, major revisions, and architectural redesigns.
4. Read the paper. Did you leave more informed or more confused? I left much more confused. Three disparate domains of computer science mashed together in a short abstract: (1) Deterministic finite-automaton equivalence, (2) Deep learning theory, (3) FPGA/VLSI Engineering. While cross domain papers exist, LLM's often mix high-level theoretical complexity bounds (like DPSACE) with hyper-specific low level vendor parameters without providing the connective tissue that bridges the math to the hardware synthesis.
5. Check the numbers. 3.81 x 10^-5 pJ/op. This is ultra precision, yet right after dropping these numbers, we get the classic "All energy figures are simulation-based tool estimates. No physical FPGA board measurement was performed". One of the earlier versions claimed a 2,129x speedup vs an RTX 3090 on bus-traffic reduction.
6. Consult real frontier LLM's. Straight up copy paste the link into "relatively" credible models like Claude, GPT, Grok, and Gemini and ask "Is this AI Slop?". In my case I did it with three separate models, and all 3 came back with a resounding yes, which confirmed my initial suspicion.
I am not claiming that the guy doesn't know what he is talking about. And no shame to the author, I too have fallen victim to grandiose claims from LLMs. I am just stating that this paper by itself is mega sus and doesn't belong in the top ten results of hackernews.
Thanks for the detailed writeup, these all make a lot of sense.
My only other tell (hadn’t committed the time to reading the rest of the paper nor looking into the author yet) so far was Zenodo, unfortunately that definitely has become a tell or at least hint when other signs of “slop-thought” are present, even in my very limited experience/exposure.
This is great! It reminds me of a more modern version of sandpack. Would love to see a roadmap for more upcoming features (hopefully support for svelte/vite/etc.).
Good news on that front: Vite and Svelte already work today. The Studio ships templates for Svelte 5 (Vite) and SvelteKit, alongside React, Vue, Astro, Nuxt, Solid, Qwik, Preact and more, and Vite is the default dev server for most of the frontend templates. If you open the Studio you can boot a Svelte + Vite project right now.
The Sandpack comparison is fair, though the approach differs a bit. Sandpack's classic setup bundles in the browser, whereas Vivari runs a real Node (plus Bun and Python) runtime with a virtual filesystem and a process model, so the actual npm dev server runs in-VM instead of a bundler standing in for it. Closer in spirit to WebContainer or CodeSandbox's Nodebox.
On the roadmap: near term I'm focused on hardening what's already there (Next.js and Postgres via PGlite are still experimental), expanding Bun coverage beyond the current shim, and improving the debugger and git support. I'll put up a proper public roadmap soon. If there's a specific framework or workflow you'd want, I'd genuinely like to hear it.
