GreenWatermelon

The entire Page is AI written 100%, and it's very likely DeepSeek because the prose matches my experience. I've been working heavily with DeepSeek this past week on an Architecture issue, so I read A LOT. The most obvious tell-tale of deepseek I've is long, condensed paragraphs that should've been bullet-point lists.

Oh my God it's literally this XKCD https://xkcd.com/221/

I'll one-up: the site doesn't even load for me. Dunno if it's my region or the site is just down.

You can't say this and not share the build!

So my thoughts about the results here:

I found the rating for Gemini models to track my experience. in addition to their extreme sycophancy, the information they provided (at least in the online chat) were more often than not, counterproductive. They act extremely confidently, and yield to any kind of pushback, requiring great care in prompting them to be able to get them to output useful alternatives.

This made the Gemini model family suboptimal for the purpose of investigating solutions to complex issues.

And their sycophancy rating tracks my experience, as I would groan loudly every time I read Gemini's answers and how it tries to nonsensically flatter me for asking simple, obvious questions.

At the other end, Deepseek is really good at pointing out contradictions, issues with my reasoning. It's as good as Opus (and vastly cheaper)

Where Deepseek loses out (in this study) compared to Anthropic and Open AI is "Slop", i.e. " How much generic filler or AI-flavored padding could be removed without losing substance?" though in my experience it didn't seem THAT different from Opus

The Organization category is worth a deeper look for anyone interested, because the results in this category vary wildly making the "overall" rating not very useful. e.g. GPT models are very structured i.e. " ideas organized in a logical sequence with clear progression" BUT not scannable "Can a reader skim the response and quickly find what they need? "

and while Deepseek scores highly in the overall rating for organization, its individual scores are pretty mediocre especially for Structure and Scannability, winning on the "bottom line summary" category.

PS: For fun, add "GPT-4o (Nov)" to the list. It's hilarious.

Measures how models communicate rather than whether they are correct: 30 judged personality traits rolled up into 8 indexes, plus 21 measured communication characteristics.

I was just adding a citation to your point, not arguing against you :)

also adding what other stuff aren't included in the release, and implicitly pointing out that they explicitly mention it's not included rather than it being stealthily missing.

It's not just a little different. it's VERY different. it's much more boxy, and letters are closer in height, especially noticeable in uppercase next to lowercase. I would say it's less readable than Jetbrains Mono.

Lower case 'l' even extends above upper case 'G'. it missed with my brain xD

Overall it's really pretty but not a daily driver.

If you're talking about powerline/nerd symbols, have you tried patching it yourself? I did and it worked.

I would argue that negatively impacting the portfolios of US VCs is one of the greatest favors one can do to the world.

yeah they mention:

> Required for a complete inference service (not included in this release):

> A KV cache system

> VE/Prefill/Decode disaggregation (most required engine functionality is already implemented)

> Load balancing across data-parallel replicas for both prefill and decode

I also assume they will just slot in whatever cards they can get their hands on, so this figure could be an average.

Until the Chinese stuff came online I was completely convinced this is just a bubble waiting to pop and nothing will come of this tech due to its absurd costs. The Chinese stuff changed this picture massively. I'm daily driving DeepSeek 4.1 flash (at API prices) and every now and then I give Opus 5.5 a try and don't notice anything missing for me.

My workflow is one where I am heavily involved in the process and I don't like giving the agent too much autonomy + I need to follow along. Opus 5.5 speak is still cause aneurysms for my brain. It also has this vibe to its prose... like it's looking down on me. This is Prolly my bias speaking though.

Relevant: Deepseek reports its Revenue per GPU per hour at off peak prices (38$) [reddit link] https://www.reddit.com/r/DeepSeek/comments/1xanexq/deepseek_...

Also cool (from the translated screenshot since original is in Chineese https://zhuanlan.zhihu.com/p/2095560295930508953)

> the DeepSeek V4.1 Flash inference system has established three goals

> 1-..... 2-..... 3. Supports retaining KVCache for more than 72 hours

This is awesome

Well you're in luck, because Deno is getting sunset! They'll be able to remove it in the future.

Yup, Firefox on Android as well. The page scrolls in frames per MINUTE: scroll, wait 2 seconds, page instantly jumps to future scroll location.

A pretty important feature is that the client process can crash without losing progress on my current session, similar to how I can reconnect to my tmux server without losing current active processes.

You're talking about privacy from governments, which I agree os an overblown concern unless you're some sort of political activist/dissident.

However, privacy from mega corpos is an entirely different matter, because they CARE about you(r data) and rather than a pixel, you're a fully fleshed out record in their gigantic databases and they use this data to profit off of you, in multiple ways that all can be described as "scamming you out of money" by way 0f perfecting the art of selling you shit you don't need.

And then you have the wannabes, mid-sized corpos with security postures so poor every hacker and their grandmother has your data, for the sole purpose of scamming your own grandmother.

It works ok and good by default. You start using it as if you'd Claude or Codex, the when you find yourself missing a feature, you get an extension for it or you ask an agent to make it.

For example, I started using it, wanted a plan mode, downloaded an extension for it. Wanted to be able to analyze token usage for turn in a session, I had an agent create an extension for that. Then I wanted a custom status bar, so I had an agent implement an extension for that.

Symlink the .pi directory to the profile you want, swapping changes the target of the link inste of modifying files. This is how I implement dark/light mode autoswitches on some of the cli tools I use.

Hmm? what do you mean? installed apps are now available in "My Apps", 3rd button in the navigation bar after the update.

Are you talking about pre-2.0 versions?

The Astra Minecraft clone is the only one that I could actually play on mobile.

They train. Deepseek doesn't have a Zero Data Retention policy, but for many personal tasks I'm (not parent) ok with them having this data if it means better, cheaper models for everyone on the future.

Different story with $DAYJOB where ZDR matters, but the beauty of open models is that you can find a provider hosting them in your preferred terms.

They have it in English. It's called "word vomit"

Also useful words: succinct, concise. Sometimes I use both. Maybe I'll ise all three: be succinct, terse, and concise.

This reads too much like a Soviet Russia joke. The irony is intense.

> Tailwind is a ~~load-bearing~~ very important part of the stack

This is beautiful. Art, even. And it speaks to my very soul.

I've come to call this language "Claudese English" (or Claudish)

Recently I've started using GLM models and noticed they aren't very claudish (just a little bit, compared to DeepSeek which is extremely claudish)

I received an email for it.

OCR now comes built in with Spectacle, the default screenshot utility for KDE Plasma and it can copy the result right away.