iiinova — where your digital self keeps evolving

Edward J. Schwartz @ejschwar.bsky.social · 8h

I'm really enjoying the "Statistical Rethinking" course/book. It should be required reading for all researchers. oceanrep.geomar.de/id/eprint/55... github.com/rmcelreath/s...

oceanrep.geomar.de

I Challenge Thee

000

Edward J. Schwartz @ejschwar.bsky.social · 17/09/2026

arena.ai/blog/coding-... This is interesting. They find that harness has little impact on success, but significant impact on cost.

arena.ai

HarnessTax: How Much Does the Harness Matter for Coding Agents? - Arena.ai

001

Edward J. Schwartz @ejschwar.bsky.social · 24/08/2026

www.latent.space/p/attention-... 1. Interesting view of AI timeline 2. I agree that models and harnesses are being intertwined, and I'm not sure it's a good thing.

latent.space

The Evolution of the Agent Harness

Models keep absorbing the harness into their weights — soon, it will be a harness for human attention rather than for the model.

000

Edward J. Schwartz @ejschwar.bsky.social · 25/07/2026

Harbor's hub now lets you upload benchmark runs, which is pretty neat. Here's an example run from the auto-bench blogs I've been writing: hub.harborframework.com/jobs/4149711... You can view each run and its trajectory: hub.harborframework.com/jobs/4149711...

hub.harborframework.com

Harbor Hub

Browse Harbor Hub

001

Reposted by Edward J. Schwartz

Dr. Claire Le Goues @clegoues.bsky.social · 15/07/2026

Congratulations to Dr Luke Dramko (my lucky 13th PhD graduate!) for his successful and stellar defense yesterday of “Neural Decompilation with Minimal Risk.” The last chapter isn’t published yet but STAY TUNED because it’s a symbolic approach for proving the fidelity of neurally-decompiled code. 1/

Luke Dramko (center) with his thesis committee and the title slide from his dissertation defense. Claire is indeed short but Luke and Rohan are also quite tall.

162

Edward J. Schwartz @ejschwar.bsky.social · 22/07/2026

🚨 Blog Post: "Benchmarking Quantized LLMs for Local Coding Agents Part 4: Investigating the Impact of Thinking on Qwen3.5-35B-A3B" edmcman.github.io/blog/2026-07-22--…

121

Edward J. Schwartz @ejschwar.bsky.social · 17/07/2026

I'm old enough that I remember legitimately using this! www.osnews.com/story/145534...

osnews.com

Microsoft releases its weird ’90s IRC client as open source – OSnews

000

Edward J. Schwartz @ejschwar.bsky.social · 04/07/2026

🚨 Blog Post: "Benchmarking Quantized LLMs for Local Coding Agents Part 3: Investigating the Performance of Qwen3.5-35B-A3B (Take 2)" edmcman.github.io/blog/2026-07-04--…

000

Edward J. Schwartz @ejschwar.bsky.social · 12/06/2026

🚨 Blog Post: "Benchmarking Quantized LLMs for Local Coding Agents Part 3: Investigating the Performance of Qwen3.5-35B-A3B" edmcman.github.io/blog/2026-06-12--…

000

Edward J. Schwartz @ejschwar.bsky.social · 04/06/2026

🚨 Blog Post: "auto-bench: Benchmarking Quantized LLMs for Local Coding Agents Part 2" edmcman.github.io/blog/2026-06-04--…

000

Edward J. Schwartz @ejschwar.bsky.social · 28/05/2026

🚨 Blog Post: "ASP: Towards the Next Generation OOAnalyzer" edmcman.github.io/blog/2026-05-28--…

000

Reposted by Edward J. Schwartz

Ryan @snarfed.org · 12/05/2026

my google account is heaven@gmail (long story), so I get a steady trickle of emails from people writing to God. some of them are real winners

Screenshot of an email

To: god, me
Subject: Arsenal

Dear God,

I know you got a lot of things to do but you need to stop what you are doing and help Arsenal beat West Ham. Pretty please 🙏

Regards
Your Best Human

18668164

Edward J. Schwartz @ejschwar.bsky.social · 15/05/2026

🚨 Blog Post: "auto-bench: Benchmarking Quantized LLMs for Local Coding Agents" edmcman.github.io/blog/2026-05-15--…

010

Edward J. Schwartz @ejschwar.bsky.social · 05/05/2026

x.com/nicbstme/sta... Models are being tuned for specific harnesses (shocker /s)

x.com

000

Edward J. Schwartz @ejschwar.bsky.social · 04/05/2026

Ubuntu is being ddosed and their repositories are down. Bad news if you need apt because you want to build a devcontainer or run your CI!

010

Edward J. Schwartz @ejschwar.bsky.social · 01/05/2026

I really wish there was a leaderboard for GGUFs on swe-bench or something similar...

000

Edward J. Schwartz @ejschwar.bsky.social · 23/03/2026

www.reddit.com/r/ClaudeAI/c... AI providers are beginning to slow down their subsidies on LLM usage. For now, it is only external agents like Opencode. But LLM access as a subscription model is a loss leader. All these vibe coders are going to be very upset when they have to pay per token.

reddit.com

From the ClaudeAI community on Reddit: Claude subscriptions will no longer be usable in Opencode.

Explore this post and more from the ClaudeAI community

110

Edward J. Schwartz @ejschwar.bsky.social · 01/03/2026

I designed my first real part! www.printables.com/model/162187... No, I don't really know what I'm doing. Yes, it is satisfying anyway!

printables.com

Case for Waveshare ESP32-C6-Geek Development Board by Edward Schwartz | Download free STL model | Printables.com

000

Edward J. Schwartz @ejschwar.bsky.social · 20/02/2026

TIL about pypi-timemachine. You're welcome. (You know, for those 10-year old research projects without lock files)

000

Edward J. Schwartz @ejschwar.bsky.social · 18/02/2026

snap: When you only care about 90% of your apps to work. It's been years. Why do major snaps (firefox!) still have usability issues?!

000

Edward J. Schwartz @ejschwar.bsky.social · 12/02/2026

openai.com/index/harnes... "What’s become clear: building software still demands discipline, but the discipline shows up more in the scaffolding rather than the code. The tooling, abstractions, and feedback loops that keep the codebase coherent are increasingly important."

openai.com

Harness engineering: leveraging Codex in an agent-first world

By Ryan Lopopolo, Member of the Technical Staff

020

Edward J. Schwartz @ejschwar.bsky.social · 11/02/2026

x.com/mrexodia/sta... I concur with most of this advice on vibe coding.

x.com

000

Edward J. Schwartz @ejschwar.bsky.social · 21/01/2026

TIL that fine-tuning a PEFT adapter for a pretrained model (i.e., not a model fine-tuned for chat) is probably a bad idea. By default, PEFT adapters don't include the vocab embeddings, so it is more or less unable to learn the meaning of tokens not in pretraining like EOS. Whoops!

000

Edward J. Schwartz @ejschwar.bsky.social · 15/01/2026

🚨 Blog Post: ""Idioms: A Simple and Effective Framework for Turbo-Charging Local Neural Decompilation with Well-Define... edmcman.github.io/blog/2026-01-15--…

011

Reposted by Edward J. Schwartz

ACM SURE Workshop @sureworkshop.bsky.social · 08/01/2026

Reflecting on the success of our first SURE and beginning the planning for the next year!

011

Edward J. Schwartz @ejschwar.bsky.social · 07/01/2026

Some days I hate computers. Today I've been debugging a packer build of windows 11 arm using vmware fusion, which is uncharted territory already. There were many problems. But the last and most frustrating was Mac silently blocking access to my "local network" bc I was in VS Code. 🤯

100

Edward J. Schwartz @ejschwar.bsky.social · 06/01/2026

Today's cool visualization of the day is brought to you by arxiv.org/pdf/2512.14045 The world needs more Sankey diagrams.

001

Edward J. Schwartz @ejschwar.bsky.social · 22/12/2025

I always felt like tqdm's ETA estimations were wildly inaccurate. It's because it defaults to using an exponentially weighted moving average with 0.3 weight. With high variance in job times and a lot of threads, that isn't going to work well. tqdm.github.io/docs/tqdm/#:....

tqdm.github.io

tqdm.tqdm - tqdm documentation

A Fast, Extensible Progress Meter

010

Edward J. Schwartz @ejschwar.bsky.social · 22/12/2025

Not a good vibe coding day. GH Co-pilot couldn't seem to figure out how to read a file. Problem 1: Task was given to sub-agent without the path. Oops. Problem 2: I had so many files in the workspace that search was timing out.

310

Edward J. Schwartz @ejschwar.bsky.social · 06/12/2025

varlogsimon.leaflet.pub/3m6zrw6k2bs2p Back in my day, TLS was considered end to end encryption. Who did you think was going to be on the other "end" of your toilet besides Kohler? Your social contacts? 🚽

varlogsimon.leaflet.pub

Kohler Can Access Data and Pictures from Toilet Camera It Describes as “End-to-End Encrypted” - /var/log/simon

Claimed end-to-end privacy doesn’t fully conceal your rear-end data

110