Reposted by Vlad Iliescu
Out of curiosity I decided to try and run the numbers on how much Netflix you can watch for the energy cost of a ChatGPT prompt
As far as I can tell it's between 5.1 and 10.2 seconds, depending on which end of the 2019 IEA Netflix energy usage estimate you use
simonwillison.net/2025/Nov/29/...
As far as I can tell it's between 5.1 and 10.2 seconds, depending on which end of the 2019 IEA Netflix energy usage estimate you use
simonwillison.net/2025/Nov/29/...
November 29, 2025 at 2:16 AM
Out of curiosity I decided to try and run the numbers on how much Netflix you can watch for the energy cost of a ChatGPT prompt
As far as I can tell it's between 5.1 and 10.2 seconds, depending on which end of the 2019 IEA Netflix energy usage estimate you use
simonwillison.net/2025/Nov/29/...
As far as I can tell it's between 5.1 and 10.2 seconds, depending on which end of the 2019 IEA Netflix energy usage estimate you use
simonwillison.net/2025/Nov/29/...
Reposted by Vlad Iliescu
Jump in puddles.
Tomorrow that water will be a cloud but now’s your chance to kick it while it’s down.
Tomorrow that water will be a cloud but now’s your chance to kick it while it’s down.
November 24, 2025 at 11:26 AM
Jump in puddles.
Tomorrow that water will be a cloud but now’s your chance to kick it while it’s down.
Tomorrow that water will be a cloud but now’s your chance to kick it while it’s down.
This could be a slide for the correlation != causation class
November 16, 2025 at 3:00 PM
This could be a slide for the correlation != causation class
Reposted by Vlad Iliescu
Americans Demand New Form Of Media To Bridge Entertainment Gap While Looking From Laptop To Phone https://theonion.com/americans-demand-new-form-of-media-to-bridge-entertainm-1819576756/
November 13, 2025 at 3:00 PM
Americans Demand New Form Of Media To Bridge Entertainment Gap While Looking From Laptop To Phone https://theonion.com/americans-demand-new-form-of-media-to-bridge-entertainm-1819576756/
Reposted by Vlad Iliescu
Can LLMs accurately aggregate information over long, information-dense texts? Not yet…
We introduce Oolong, a dataset of simple-to-verify information aggregation questions over long inputs. No model achieves >50% accuracy at 128K on Oolong!
We introduce Oolong, a dataset of simple-to-verify information aggregation questions over long inputs. No model achieves >50% accuracy at 128K on Oolong!
November 7, 2025 at 5:07 PM
Can LLMs accurately aggregate information over long, information-dense texts? Not yet…
We introduce Oolong, a dataset of simple-to-verify information aggregation questions over long inputs. No model achieves >50% accuracy at 128K on Oolong!
We introduce Oolong, a dataset of simple-to-verify information aggregation questions over long inputs. No model achieves >50% accuracy at 128K on Oolong!
Reposted by Vlad Iliescu
The Sleaze Theorem: Any technology that can reshape society will be applied to sleazy ends first.
Examples: credit cards and porn, optical storage and porn, synthetic voice generation and scam calls, email and scams/porn, the internet and ...
This has implications for likely AI x-risk scenarios 1/
Examples: credit cards and porn, optical storage and porn, synthetic voice generation and scam calls, email and scams/porn, the internet and ...
This has implications for likely AI x-risk scenarios 1/
November 21, 2023 at 4:47 PM
The Sleaze Theorem: Any technology that can reshape society will be applied to sleazy ends first.
Examples: credit cards and porn, optical storage and porn, synthetic voice generation and scam calls, email and scams/porn, the internet and ...
This has implications for likely AI x-risk scenarios 1/
Examples: credit cards and porn, optical storage and porn, synthetic voice generation and scam calls, email and scams/porn, the internet and ...
This has implications for likely AI x-risk scenarios 1/
Reposted by Vlad Iliescu
"A surgeon isn't a manager, they do the actual work! But their skills and time are highly leveraged with a support team that handles prep, secondary tasks, admin. The surgeon focuses on the important stuff they are uniquely good at."
October 24, 2025 at 2:29 PM
"A surgeon isn't a manager, they do the actual work! But their skills and time are highly leveraged with a support team that handles prep, secondary tasks, admin. The surgeon focuses on the important stuff they are uniquely good at."
Zed is cool and all, but I'm getting real tired of this shit. Wasn't even using the thing, just made the mistake of opening up two instances and leaving them in the background for a couple hours. Now I'm all out of battery 🤦♂️
October 17, 2025 at 5:11 PM
Zed is cool and all, but I'm getting real tired of this shit. Wasn't even using the thing, just made the mistake of opening up two instances and leaving them in the background for a couple hours. Now I'm all out of battery 🤦♂️
Reposted by Vlad Iliescu
"I have a decent fluency in LLMs, and they have utility, but the absurd degree of over-hype, the way they're being forced on everyone, and the insistence on ignoring the many valid critiques about them make it very difficult to focus on legitimate uses where they might add value."
October 17, 2025 at 4:32 AM
"I have a decent fluency in LLMs, and they have utility, but the absurd degree of over-hype, the way they're being forced on everyone, and the insistence on ignoring the many valid critiques about them make it very difficult to focus on legitimate uses where they might add value."
Reposted by Vlad Iliescu
Vibe coding is irresponsibly building software through dice rolls, not caring what code is produced
What about when engineers at the top of their game use AI tools responsibly to accelerate their work?
I propose "vibe engineering"!
simonwillison.net/2025/Oct/7/v...
What about when engineers at the top of their game use AI tools responsibly to accelerate their work?
I propose "vibe engineering"!
simonwillison.net/2025/Oct/7/v...
Vibe engineering
I feel like vibe coding is pretty well established now as covering the fast, loose and irresponsible way of building software with AI—entirely prompt-driven, and with no attention paid to …
simonwillison.net
October 7, 2025 at 2:35 PM
Vibe coding is irresponsibly building software through dice rolls, not caring what code is produced
What about when engineers at the top of their game use AI tools responsibly to accelerate their work?
I propose "vibe engineering"!
simonwillison.net/2025/Oct/7/v...
What about when engineers at the top of their game use AI tools responsibly to accelerate their work?
I propose "vibe engineering"!
simonwillison.net/2025/Oct/7/v...
Reposted by Vlad Iliescu
When you try to predict about the impact of AI tools on software engineering - a great analogy I heard from @simonwillison.net is to look at the impact of smartphones (+ the amazing cameras they have) on filmmaking.
Noticed that those able to shoot pro films using a smartphone are... pros?
Noticed that those able to shoot pro films using a smartphone are... pros?
October 2, 2025 at 8:26 AM
When you try to predict about the impact of AI tools on software engineering - a great analogy I heard from @simonwillison.net is to look at the impact of smartphones (+ the amazing cameras they have) on filmmaking.
Noticed that those able to shoot pro films using a smartphone are... pros?
Noticed that those able to shoot pro films using a smartphone are... pros?
Reposted by Vlad Iliescu
ok this breakdown of AI writing indicators is incredible. Huge shout-out to the Wikipedia community for this
Wikipedia:Signs of AI writing - Wikipedia
en.wikipedia.org
September 28, 2025 at 2:41 PM
ok this breakdown of AI writing indicators is incredible. Huge shout-out to the Wikipedia community for this
Reposted by Vlad Iliescu
Q: Why did the LLM cross the road?
A: We're not sure, but it achieved 94.7% on CHIKENBench-Large
A: We're not sure, but it achieved 94.7% on CHIKENBench-Large
September 29, 2025 at 1:00 PM
Q: Why did the LLM cross the road?
A: We're not sure, but it achieved 94.7% on CHIKENBench-Large
A: We're not sure, but it achieved 94.7% on CHIKENBench-Large
iTerm2 has released their AI Chat feature, so I've written a quick guide on how to set it up with Azure AI, which is useful if you want some sort of a safety net against prompt injection + data leakage.
Available here -- vladiliescu.net/iterm2-with-...
Available here -- vladiliescu.net/iterm2-with-...
Using iTerm2's new AI Chat features with Azure OpenAI
Configure iTerm2’s AI Chat to use a private Azure OpenAI deployment with guardrails, minimal permissions, and a few gotchas to avoid.
vladiliescu.net
September 26, 2025 at 11:08 AM
iTerm2 has released their AI Chat feature, so I've written a quick guide on how to set it up with Azure AI, which is useful if you want some sort of a safety net against prompt injection + data leakage.
Available here -- vladiliescu.net/iterm2-with-...
Available here -- vladiliescu.net/iterm2-with-...
Interesting postmortem from Anthropic of the issues plaguing Claude this summer. Plus there's a tiny detail, hidden deep within the article, that might explain their sudden 180-degree turn on privacy:
www.anthropic.com/engineering/...
www.anthropic.com/engineering/...
A postmortem of three recent issues
This is a technical report on three bugs that intermittently degraded responses from Claude. Below we explain what happened, why it took time to fix, and what we're changing.
www.anthropic.com
September 18, 2025 at 9:38 AM
Interesting postmortem from Anthropic of the issues plaguing Claude this summer. Plus there's a tiny detail, hidden deep within the article, that might explain their sudden 180-degree turn on privacy:
www.anthropic.com/engineering/...
www.anthropic.com/engineering/...
Reposted by Vlad Iliescu
Roses are red
Our resolve hasn't faltered
Our resolve hasn't faltered
February 14, 2025 at 1:38 PM
Roses are red
Our resolve hasn't faltered
Our resolve hasn't faltered
Reposted by Vlad Iliescu
I don't really care if the future is short videos. I like words. Words have gotten me this far and I'm gonna stick with 'em for the duration
September 9, 2025 at 6:21 PM
I don't really care if the future is short videos. I like words. Words have gotten me this far and I'm gonna stick with 'em for the duration
Reposted by Vlad Iliescu
Interesting observation from an eng manager:
“As soon as I know some text is AI-generated: I lose all interest in reading it.
For performance reviews, I asked people to either not use AI or if they must: just write down the prompt so I don’t need to go thru the generated word salad padding.”
“As soon as I know some text is AI-generated: I lose all interest in reading it.
For performance reviews, I asked people to either not use AI or if they must: just write down the prompt so I don’t need to go thru the generated word salad padding.”
September 3, 2025 at 12:26 PM
Interesting observation from an eng manager:
“As soon as I know some text is AI-generated: I lose all interest in reading it.
For performance reviews, I asked people to either not use AI or if they must: just write down the prompt so I don’t need to go thru the generated word salad padding.”
“As soon as I know some text is AI-generated: I lose all interest in reading it.
For performance reviews, I asked people to either not use AI or if they must: just write down the prompt so I don’t need to go thru the generated word salad padding.”
Your coding agent implementation matters just as much as the underlying model.
I've just spend 15 minutes trying to cajole GitHub Copilot (both Claude Sonnet 4 and GPT-5) to implement a rather exotic change in my codebase.
I've just spend 15 minutes trying to cajole GitHub Copilot (both Claude Sonnet 4 and GPT-5) to implement a rather exotic change in my codebase.
September 8, 2025 at 9:47 AM
Your coding agent implementation matters just as much as the underlying model.
I've just spend 15 minutes trying to cajole GitHub Copilot (both Claude Sonnet 4 and GPT-5) to implement a rather exotic change in my codebase.
I've just spend 15 minutes trying to cajole GitHub Copilot (both Claude Sonnet 4 and GPT-5) to implement a rather exotic change in my codebase.
Reposted by Vlad Iliescu
About two-thirds of patients who were flagged by the AI stethoscope as potentially having heart failure did not actually have it.
AI stethoscope can detect heart issues in 15 seconds, UK doctors find
About two-thirds of patients who were flagged by the AI stethoscope as potentially having heart failure did not actually have it.
l.euronews.com
September 1, 2025 at 11:05 PM
About two-thirds of patients who were flagged by the AI stethoscope as potentially having heart failure did not actually have it.
Reposted by Vlad Iliescu
Too many quotes to pull from this piece by @pedramnavid.com, but "The cost of thinking has been offloaded to the consumer, not the producer" really hits databased.pedramnavid.com/p/the-cost-o...
The Cost of Costing Nothing
The emotional labor of reading your Chat
databased.pedramnavid.com
August 5, 2025 at 4:50 PM
Too many quotes to pull from this piece by @pedramnavid.com, but "The cost of thinking has been offloaded to the consumer, not the producer" really hits databased.pedramnavid.com/p/the-cost-o...
Reposted by Vlad Iliescu
Google Gemini’s head of product says PMs are ditching PRDs for vibe coding prototypes.
But PRDs define what you’re building and why, priorities, success metrics and its features while prototypes only show its features.
Skipping these is risky as teams may ship without clarity on goals or tradeoffs
But PRDs define what you’re building and why, priorities, success metrics and its features while prototypes only show its features.
Skipping these is risky as teams may ship without clarity on goals or tradeoffs
August 17, 2025 at 12:36 PM
Google Gemini’s head of product says PMs are ditching PRDs for vibe coding prototypes.
But PRDs define what you’re building and why, priorities, success metrics and its features while prototypes only show its features.
Skipping these is risky as teams may ship without clarity on goals or tradeoffs
But PRDs define what you’re building and why, priorities, success metrics and its features while prototypes only show its features.
Skipping these is risky as teams may ship without clarity on goals or tradeoffs
Reposted by Vlad Iliescu
The first time Charlie Chaplin used his voice in a film, he wielded it to forcefully condemn the rising forces of fascism around the globe. Will we listen to his words today?
August 13, 2025 at 10:29 PM
The first time Charlie Chaplin used his voice in a film, he wielded it to forcefully condemn the rising forces of fascism around the globe. Will we listen to his words today?
Reposted by Vlad Iliescu
Vote for the 2025 Tiny Awards Site of the Year! This looks like a fantastic group of nominees that celebrate “the best of the small, poetic, creative, handmade web”. [tinyawards.net]
Tiny Awards
This is the home of the Tiny Awards, which, since 2023, has celebrated the best of the small, poetic, creative, handmade web.
tinyawards.net
August 12, 2025 at 2:52 PM
Vote for the 2025 Tiny Awards Site of the Year! This looks like a fantastic group of nominees that celebrate “the best of the small, poetic, creative, handmade web”. [tinyawards.net]
Reposted by Vlad Iliescu
It’s just matrix multiplication
August 10, 2025 at 1:56 AM
It’s just matrix multiplication