evan king
@evanking.io
AI/ML researcher working with tiny audio, speech, and language models. occasionally exploring the intersection of code, ML, and design.
ece phd @ ut austin, 2024
cs bs @ unm, 2017
opinions are my own.
github.com/evmaki
ece phd @ ut austin, 2024
cs bs @ unm, 2017
opinions are my own.
github.com/evmaki
Two qualitative observations from playing with agents:
1. They're only as good as their initial conditions. Good instructions ("SOUL.md") produce pretty thoughtful agents; bad ones produce slop generators.
2. There's a lower bound to slop. Even "perfect" instructions produce a faint whiff of slop.
1. They're only as good as their initial conditions. Good instructions ("SOUL.md") produce pretty thoughtful agents; bad ones produce slop generators.
2. There's a lower bound to slop. Even "perfect" instructions produce a faint whiff of slop.
February 13, 2026 at 8:42 PM
Two qualitative observations from playing with agents:
1. They're only as good as their initial conditions. Good instructions ("SOUL.md") produce pretty thoughtful agents; bad ones produce slop generators.
2. There's a lower bound to slop. Even "perfect" instructions produce a faint whiff of slop.
1. They're only as good as their initial conditions. Good instructions ("SOUL.md") produce pretty thoughtful agents; bad ones produce slop generators.
2. There's a lower bound to slop. Even "perfect" instructions produce a faint whiff of slop.
That t-SNE of the different Bluesky communities making the rounds is pretty cool, but so is this live 3D visualization of the firehose. Guess I missed it the first time around? Kinda relaxing to lean back and let it wash over you @theo.io
firehose3d.theo.io
firehose3d.theo.io
Bluesky Firehose in 3D (live)
firehose3d.theo.io
February 10, 2026 at 6:46 AM
That t-SNE of the different Bluesky communities making the rounds is pretty cool, but so is this live 3D visualization of the firehose. Guess I missed it the first time around? Kinda relaxing to lean back and let it wash over you @theo.io
firehose3d.theo.io
firehose3d.theo.io
Finishing up evals for the new Moonshine v2 speech-to-text models we're rolling out, and it's... exciting. We're right there with NVIDIA pushing the accuracy frontier while doing it with much smaller models, a team of < 10 engineers, and microscopic GPU spend.
February 9, 2026 at 11:38 PM
Finishing up evals for the new Moonshine v2 speech-to-text models we're rolling out, and it's... exciting. We're right there with NVIDIA pushing the accuracy frontier while doing it with much smaller models, a team of < 10 engineers, and microscopic GPU spend.
Had a great time chatting career trajectories, on-device AI, and model specialization versus generalization with @petewarden.bsky.social on the @eetimes.bsky.social Silicon Grapevine podcast
youtu.be/XbSQtaZ_QM0?...
youtu.be/XbSQtaZ_QM0?...
Edge to Impact: Building Human-Centric AI Models at Scale
YouTube video by EE Times
youtu.be
February 6, 2026 at 11:27 PM
Had a great time chatting career trajectories, on-device AI, and model specialization versus generalization with @petewarden.bsky.social on the @eetimes.bsky.social Silicon Grapevine podcast
youtu.be/XbSQtaZ_QM0?...
youtu.be/XbSQtaZ_QM0?...
Excited to start sharing more about the Mirage, a generative groovebox sampler I've been developing these past few months. It uses tiny, on-device generative audio models and a voice interface (powered by Moonshine models) to create new sounds on the fly.
February 1, 2026 at 6:56 PM
Excited to start sharing more about the Mirage, a generative groovebox sampler I've been developing these past few months. It uses tiny, on-device generative audio models and a voice interface (powered by Moonshine models) to create new sounds on the fly.
An early (~2010) project I never finished was a box that detected nearby weather and generated appropriate ambient music. This was pre-RPi, before I knew of Eno, tools like Wotja/Pd, etc. I started from the ground up (in Java, no less) and it was too much work. Now it could be vibecoded in a day 🤔
January 27, 2026 at 7:17 PM
An early (~2010) project I never finished was a box that detected nearby weather and generated appropriate ambient music. This was pre-RPi, before I knew of Eno, tools like Wotja/Pd, etc. I started from the ground up (in Java, no less) and it was too much work. Now it could be vibecoded in a day 🤔
This is a positive thing as far as the historical record is concerned. It will preserve the names and likenesses of people that were brave enough to stand up for their community.
TOM HOMAN: “We’re going to create a database where those people that are arrested for interference, impeding and assault, we’re going to make them famous. We’re going to put their face on TV. We’re going to let their employers, in their neighborhoods, in their schools, know who these people are.”
Story: www.cnn.com/2026/01/27/u...
January 27, 2026 at 7:05 PM
This is a positive thing as far as the historical record is concerned. It will preserve the names and likenesses of people that were brave enough to stand up for their community.
Work in tech often rewards thinking only in abstractions, which makes it easy to miss downstream impacts once your projects are shipped out the door. Building stronger ethical literacy feels increasingly important.
SCOOP: Palantir Defends Work With ICE to Staff Following Killing of Alex Pretti
WIRED obtained Slack conversations + an updated internal Palantir wiki defending the company's work for ICE to outraged workers.
More here:
www.wired.com/story/palant...
WIRED obtained Slack conversations + an updated internal Palantir wiki defending the company's work for ICE to outraged workers.
More here:
www.wired.com/story/palant...
Palantir Defends Work With ICE to Staff Following Killing of Alex Pretti
“In my opinion ICE are the bad guys. I am not proud that the company I enjoy so much working for is part of this,” one worker wrote on Slack.
www.wired.com
January 27, 2026 at 1:19 AM
Work in tech often rewards thinking only in abstractions, which makes it easy to miss downstream impacts once your projects are shipped out the door. Building stronger ethical literacy feels increasingly important.
Another strong case against the idea that everything AI should be cloud-connected. The privacy liability vanishes when you do it on-device, at the edge.
Google has agreed to pay $68 million to settle a class-action lawsuit that alleged the technology giant's voice assistant had illegally recorded users and then shared their private conversations with advertisers.
Google to pay $68 million over allegations its voice assistant eavesdropped on users
Class-action lawsuit alleged that Google's voice assistant illegally recorded and shared private conversations with advertisers.
cbsn.ws
January 26, 2026 at 9:39 PM
Another strong case against the idea that everything AI should be cloud-connected. The privacy liability vanishes when you do it on-device, at the edge.
It’s easy to forget that global political powers have been engaged in continuous information warfare for 10+ years. Many mistake the radical and disgusting takes they see online for what “real” people think. Even among real people online there is a selection bias – normal people don’t post all day.
January 25, 2026 at 5:25 PM
It’s easy to forget that global political powers have been engaged in continuous information warfare for 10+ years. Many mistake the radical and disgusting takes they see online for what “real” people think. Even among real people online there is a selection bias – normal people don’t post all day.
I rarely wade into politics, but: Pretti was a model American and, importantly, a model of healthy masculinity. He protected and healed others, was in touch with nature and animals. He was executed by a federal government driven by toxic masculinity: cowardly, shameless, and self-obsessed.
January 24, 2026 at 10:47 PM
I rarely wade into politics, but: Pretti was a model American and, importantly, a model of healthy masculinity. He protected and healed others, was in touch with nature and animals. He was executed by a federal government driven by toxic masculinity: cowardly, shameless, and self-obsessed.
Now that everyone is walking around saying "it's not X, it's Y", it's obvious that AI speak has been internalized by the culture
January 22, 2026 at 9:25 PM
Now that everyone is walking around saying "it's not X, it's Y", it's obvious that AI speak has been internalized by the culture
Going from Rev 1 (left, mid December) to Rev 2 (right, mid January) of my hardware project took some long nights and a lot of soldering, but it’s really coming together.
January 20, 2026 at 10:01 AM
Going from Rev 1 (left, mid December) to Rev 2 (right, mid January) of my hardware project took some long nights and a lot of soldering, but it’s really coming together.
Many AI boosters are forgetting that the most important part of work is the subjective, human aspect: empathizing with customer needs, communicating effectively, and understanding tastes and trends. A human can prompt an LLM into effectiveness in these areas, but it will rarely get there on its own.
December 27, 2025 at 6:18 PM
Many AI boosters are forgetting that the most important part of work is the subjective, human aspect: empathizing with customer needs, communicating effectively, and understanding tastes and trends. A human can prompt an LLM into effectiveness in these areas, but it will rarely get there on its own.
Feeling a sense of relief when the AI coding output is garbage; it’s a good heuristic for the novelty of the work. But using the AI to do today’s novel work will create training data for the next model iteration 🤔
December 4, 2025 at 7:43 PM
Feeling a sense of relief when the AI coding output is garbage; it’s a good heuristic for the novelty of the work. But using the AI to do today’s novel work will create training data for the next model iteration 🤔
Keyswitch and LED circuit coming together for a step sequencer. Using a dedicated PWM driver IC for controllable brightness gives much better results than my first attempt (bitbanging a constant current LED driver in software)
December 1, 2025 at 9:30 AM
Keyswitch and LED circuit coming together for a step sequencer. Using a dedicated PWM driver IC for controllable brightness gives much better results than my first attempt (bitbanging a constant current LED driver in software)
Realized something recently: MX-style keyboard switches are great general-purpose buttons. You can experiment with different 3d printed keycaps, and they're super tactile. I just wish they were a little easier to breadboard.
November 28, 2025 at 10:13 PM
Realized something recently: MX-style keyboard switches are great general-purpose buttons. You can experiment with different 3d printed keycaps, and they're super tactile. I just wish they were a little easier to breadboard.
Still working out my relationship with coding agents. When you enjoy the process of creating, it's hard to see the value of the process diminished. You also become less capable if you offload the cognitive lifting. But they can help you overcome blockers, and learn things in your own way.
November 24, 2025 at 7:39 PM
Still working out my relationship with coding agents. When you enjoy the process of creating, it's hard to see the value of the process diminished. You also become less capable if you offload the cognitive lifting. But they can help you overcome blockers, and learn things in your own way.
Having fun building the UI for my generative sampler project. These tiny OLED displays are pretty expressive for their size
November 23, 2025 at 6:16 PM
Having fun building the UI for my generative sampler project. These tiny OLED displays are pretty expressive for their size
I was daydreaming about a scuba trip to SE Asia recently and this workflow was useful:
1. Ask for a list of dive spots with descriptions
2. Reformat into .kml (importable in Google Earth)
3. Ask for another list of tourist destinations, also export as .kml
4. Self-plan itinerary by proximity
1. Ask for a list of dive spots with descriptions
2. Reformat into .kml (importable in Google Earth)
3. Ask for another list of tourist destinations, also export as .kml
4. Self-plan itinerary by proximity
Google’s AI Mode can now help you visualize your travel plans
Google’s AI Mode can now help you visualize your travel plans
An AI mode addition for next summer’s getaway.
buff.ly
November 17, 2025 at 6:42 PM
I was daydreaming about a scuba trip to SE Asia recently and this workflow was useful:
1. Ask for a list of dive spots with descriptions
2. Reformat into .kml (importable in Google Earth)
3. Ask for another list of tourist destinations, also export as .kml
4. Self-plan itinerary by proximity
1. Ask for a list of dive spots with descriptions
2. Reformat into .kml (importable in Google Earth)
3. Ask for another list of tourist destinations, also export as .kml
4. Self-plan itinerary by proximity
Trying to cram 9 buttons, 2 pots, 8 leds, 8 resistors, 2 ICs, and an OLED display on a Pi HAT proto board
November 16, 2025 at 10:34 PM
Trying to cram 9 buttons, 2 pots, 8 leds, 8 resistors, 2 ICs, and an OLED display on a Pi HAT proto board
Adolf Loos describing planned obsolescence in 1908 is a good reminder that there's nothing new under the sun.
November 16, 2025 at 5:50 PM
Adolf Loos describing planned obsolescence in 1908 is a good reminder that there's nothing new under the sun.
AI begets a 21st century Dadaism; it shows you the form of familiar things, but they flow into one another in a situationless manner, devoid of context.
AI researchers would call this a "world model". Right now AI can approximate the distribution of the form, but has no model of the function.
AI researchers would call this a "world model". Right now AI can approximate the distribution of the form, but has no model of the function.
Oh my god some fucking tech dingdong posted this on Twitter with the caption "AI games are going to be amazing" totally seriously, you have to watch it. You have to. In full screen.
October 24, 2025 at 5:45 PM
AI begets a 21st century Dadaism; it shows you the form of familiar things, but they flow into one another in a situationless manner, devoid of context.
AI researchers would call this a "world model". Right now AI can approximate the distribution of the form, but has no model of the function.
AI researchers would call this a "world model". Right now AI can approximate the distribution of the form, but has no model of the function.
One of those rare things that tempts you to dust off the VR headset
This is rather cool, it's an endless procedurally generated museum based on Wikipedia. Nice to wander round, particularly in VR. Spent a good while in the surprisingly vast hall of soup earlier.
https://mayeclair.itch.io/museum-of-all-things
https://mayeclair.itch.io/museum-of-all-things
Museum of All Things by Maya Claire
An infinite virtual museum generated from wikipedia
mayeclair.itch.io
June 24, 2025 at 2:01 AM
One of those rare things that tempts you to dust off the VR headset
cool project: www.flowstate.cc lets you filter your feeds based on the types of content you do/don't want to see. e.g., no public freakouts, no rage bait. tools like this could help people regain agency over their digital experiences
Flowstate
Building web experiences to reverse mindless scrolling /
www.flowstate.cc
May 22, 2025 at 11:28 PM
cool project: www.flowstate.cc lets you filter your feeds based on the types of content you do/don't want to see. e.g., no public freakouts, no rage bait. tools like this could help people regain agency over their digital experiences