easygenes

I’ve used those and they are a bit fiddly. The best tool I’ve found are the needle pointed angled precision tweezers (like they mostly sell for electronics work).

Other benchmark aggregates are less favorable to GPT-OSS-120B: https://arxiv.org/abs/2508.12461

Something I was doing informally that seems very effective is asking for details about smaller cities and towns and lesser points of interest around the world. Bigger models tend to have a much better understanding and knowledge base for the more obscure places.

I think this is good for people who only have occasional need for scheduling. I’ve used cron enough that I appreciate its compactness and would find this cumbersome and difficult.

Just thought of a few others:

  Onyx - Geometry
  Verve - Street Level
  Campos - Superior Blend
  Proud Mary - Humbler
  Five Elephant - House Blend

Geometry if you want to go for a bit lighter roast that lets some berry sparkle in. Others are more riffing on the theme.

If you care for a similar cup profile with higher quality beans, I’d characterize Dunkin original blend as a medium-light roast (lighter than Starbucks’ Pike). Aroma: toasted cereal, brown sugar, roast almond, and hint of citrus. Flavor: caramel sugar, roast nut, hint of milk choc and citrus. Acidity: sweet/malic, soft and smooth. Body: light/medium, silky. Overall: Very well balanced but simplistic.

A blend to achieve this cup would rely on primarily (50-60%) Brazilian naturals for the core sweetness, with Colombian washed for the clean citric-malic acidity and silky body (~25%), and some mix of Guatemala/Honduras and Ethiopia washed for the last quarter to tweak body and acids.

If this profile (sweet-nutty, caramel-driven profiles with gentle citrus lift) is absolutely your jam, I’d check out the following blends for a similar cup put together from higher graded green beans:

  Counter Culture - Big Trouble
  Stumptown - Holler Mountain
  Onyx - Monarch
  Intelligentsia - El Gallo 
  Blue Bottle - Three Africas
  La Colombe - Nizza
  Square Mile - Red Brick Espresso
  Coffee Collective - Espresso Blend

Counter Culture and Stumptown are probably closest in profile. Intelligentsia and Blue Bottle will add more sparkle to the finish. La Colombe and Square Mile lean sweeter.

Nice app! Quick note of first thing I noticed: Would be nice to have a more detailed option for roast. In my case I have a DiFluid Omni and use it for roast analysis, so would like to be able to enter the data from it.

You're always limited by the knowledge gaps of the underlying LLM. The method is otherwise the most coherent way to work to the strengths of the LLM through disciplined context and role management. Nothing about it is UI focused, and leans more on general agile team structures than anything else.

I think none of these offer much useful insight beyond the overarching idea of peer programming beating just vibe coding.

The best structure I've found which leverages this idea is called BMAD, and treats the LLM as though it were a whole development team in an orchestrated way that you have full control over.

https://youtu.be/E_QJ8j74U_0 https://github.com/bmadcode/BMAD-METHOD

Mini batch and streaming GD make the benefits obvious and trivial. Closed form OLS is unbeatable so long as samples * params^2 is comfortably sitting in memory. You often lose that as soon as your p approaches 10^5, which is common these days. Soon as you need distributed, streaming, or your data is too tall and or too wide then first order methods are the point of call.

Ah, we’re resorting to ad machinum today. :)

CMU lecture notes [0] I think approach it in an intuitive way, starting from the Gaussian noise linear model, deriving log-likelihood, and presenting the analytic approach. Misses the bridge to gradient methods though.

For gradients, Stanford CS229 [1] jumps right into it.

[0] https://www.stat.cmu.edu/~cshalizi/mreg/15/lectures/06/lectu...

[1] https://cs229.stanford.edu/lectures-spring2022/main_notes.pd...

Yeah. Squared error is optimal when the noise is Gaussian because it estimates the conditional mean; absolute error is optimal under Laplace noise because it estimates the conditional median. If your housing data have a few eight-figure outliers, the heavy tails break the Gaussian assumption, so a full quantile regression for, say, the 90th percentile—will predict prices more robustly than plain least squares.

OLS is a straightforward way to introduce GD, and although an analytic solution exists it becomes memory and IO bound at sufficient scale, so GD is still a practical option.

This is very light and approachable but stops short of building the statistical intuition you want here. They fixate on the smoothness of squared errors without connecting that to the gaussian noise model and establishing how that relates to the predictive power against natural sorts of data.

Super-Earth Defense Ministry Broadcast: Special Bulletin

https://chatgpt.com/share/681c31e8-67f8-8011-a4b0-2bed9d4da7...

Ygg doesn't operate strictly as a shortest path protocol. It prefers lowest latency links already and they're roadmapped to handle link cost / dynamic weighting / link congestion. They're targeting a blended weighted metric overlay which would certainly prefer the chain of quick/fat links.

If this is interesting to you, there's also a new edge mesh protocol which would dovetail into it well called MeshCore. MeshCore is LoRa and MCU focused, works at L2. Ygg is L3 and could run at e.g. your LoRa SBC gateway to provide seamless global distribution of edge mesh chats and sensor relays.

https://github.com/ripplebiz/MeshCore https://youtu.be/fNWf0Mh2fJw https://chatgpt.com/share/681c281f-0a24-8011-8ec9-6d58ce3db0...

Yes, DuckDB has a postgres extension which works via libpq.

Other extensions include but aren't limited to: MySQL, AWS, Azure, Excel, Iceberg, JSON

https://duckdb.org/docs/stable/extensions/postgres.html

Amusingly, when I asked o3 to propose changes to the education system which address the author's complaints wrt writing assignments, one of the first things it suggested was transparent prompt logging (basically what the author proposes).

https://chatgpt.com/share/6817fe76-973c-8011-acf3-ef3138c144...

For about a year Hono has had mature RPC. So bhvr has a type-safe RPC story without tRPC: export your AppType, import it in the browser, and hono/client gives you fully-inferred calls plus Zod-backed validation — in a ~3 kB helper instead of tRPC’s heavier client and adapters.

You still won’t get auto-generated hooks or WebSocket subscriptions, but for plain REST/JSON the old “hand-rolled fetch” drawback is gone, leaving only edge cases like hooks, real-time, and IDE scalability as the remaining trade-offs.

bhvr is the opposite end of the spectrum from the Big Three: Next/Rails/Django hand you batteries, bhvr hands you a rocket fuel can. One Bun binary replaces npm + Node + Jest, Hono routes run everywhere from Cloudflare Workers to bare-metal Bun, and a shared types/ folder gives you end-to-end TS safety—zero extra runtime, zero config. You lose turnkey SSR, auth, and ORM magic, but if you care more about installing, hot-reloading, and deploying in under 60 seconds than about convention-heavy kitchensinks, bhvr is the leanest full-stack starter in 2025.

There’s a request code button here: https://www.catalyzex.com/paper/pygraph-robust-compiler-supp...

The average Joe with an io domain is probably a developer happier to code up a migration than be extorted so dearly.

Hate to say, but might actually be a legitimate use case for blockchain here. Identity provider which is responsible for being a source of truth on aliveness tied to a smart contract for paying annual registrar fees.

Though the traditional way would just be finding a registrar which can direct debit (e.g. CSC Global or MarkMonitor) or setting up a trust account for someone to manage it for you. Or just power of attorney plus escrowed account.

The main point of the post is to say that it is getting maintained in a fork now, to some degree. It's still a very rudimentary implementation though.

Why in the world am I getting downvoted for this? For the Claude summary? I also checked the repository manually, and agree with the summary. One of the prime use cases for LLMs is summarization.

Have you compared the experience to using Sidebery? Every FF alternative I've tried comes up short to the power of what Sidebery can do with tree style tabs.

It's beyond that. The CPU designers aren't designing for ACPI or to target the ARM PC-BSA (Base System Architecture for PCs).

Near-perfect sRGB coverage is relatively common, especially with OLEDs. Now the concern is AdobeRBG, DCI-P3, and Rec.2020 coverage* -- which are still generally lagging (basically nothing short of some $$$$$ laser projectors are anywhere close to full coverage of Rec.2020, which covers ~75% of all visible colors and is an ambitious forward-looking colorspace) .

  * sRGB < AdobeRGB ~= DCI-P3 < Rec.2020