ddxv

I'm disappointed these models continue to be closed source and so expensive.

Open weight models being 10x or more cheaper is just so much more of an unlock than incremental gains for me.

LLMs remind me a lot of spellcheck and diffusion image generation models remind me of clip art. Clip art seemed like a pretty big deal in the 90s, and slowly seemed to go away, though I'm not sure why. Maybe it was the rise of Photoshop and Indesign editing tools.

Either way, I definitely love hate the ChatGPT flyers. They will at some point be looked back on as a dated ugly format, and later as a curiosity and finally someone will collect and save them years from now.

A year ago I wasn't using Deepseek. Now I am. I guess what changed is which models people are using most for coding.

Anthropic is actively manufacturing scarcity of it's models to drive up their prices and prevent harder benchmarks showing how close the open weight models are to parity.

Hmm, I'm surprised it would be so much delayed. The current open weight models already feel mostly caught up.

Additionally, I didn't see mention of China's much cheaper electricity and subsidies for companies to build out on Huawei. It will be interesting to look back in six months to see how many of the h200s of their allotted amount do end up being purchased.

When Yeeun Jo, a student at University of Illinois at Urbana-Champaign (UIUC) contacted me in 2025 to ask about data tracking in app advertisements related to women's health and pregnancy I was a bit skeptical.

Given the FTC’s subsequent 2021 crack-down specifically targeting Flo Health for passing intimate logging metrics to Facebook and Google. I thought it was unlikely they'd find much.

Well, it's a year later and I was wrong.

Jo & Reaves found mobile apps and mobile ad networks were still tracking health data.

Their paper shows the high specificity which these events are tracked ...

Rest of my blog post & AppGoblin links (to see which apps using which ad networks currently) to the apps in paper:

https://jamesoclaire.com/2026/06/30/the-pregnancy-and-health...

I believe the comment you replied to was talking about the cost on providers like OpenCode vs Deepseek API. Deepseek API is even cheaper than the other providers for the same deepseek models.

OpenAI, though they seem to backtrack it lately, have been slowly pushing forward of their launch of ads which would be a supplemental way to support cheaper use of their models. This is currently not as great a fit as the modern day banner ads, but it will be interesting to see where they go with that.

True, outside of the dark tactics I imagined in the article, they will have to compete at lower costs. It's just that the current iteration does not feel cost competitive yet.

Slack / Discord / IRC etc

I think finding these communities and having a proper dialogue about your product is pretty useful.

Deepseek models seem as good and are like 20x cheaper for me.

I think using open weight models will solve this. I believe they are nearly caught up and much of the gains are in the harnesses or properly orchestration of subqueries. (I'm no expert, just my opinion).

When the open weight models catch up, if they don't get lobbied and banned by OpenAi and Anthropic, then you'll be able to use them to properly secure your software.

I feel like Framework really missed this boat with their desktop PC they released last year. They could have used that + a gamepad to have captured so much of this.

"Enough people have asked me about the Peter Thiel-Dialog story that I think it's worth saying what it is, or at least what I saw it to be. So:

–Dialog is a conference. I went once in 2018 and once in 2022. No one ever asked me to keep it or my presence a secret.

–My understanding was Thiel was one of its founders but no longer involved by the time I went. I never saw or talked to him in connection with Dialog.

–Nor did I see the other names I’ve heard mentioned, like Ted Cruz or Elon Musk or Joseph Gordon-Levitt or Jared Kushner. Dialog was not sold to me as a bunch of big names, which is part of why I went. I don’t need to go to a conference to hear what Ted Cruz thinks.

–You could be a Dialog member, but I wasn’t. I don’t think joining got you much except guaranteed invitations to future Dialogs. There were occasional dinners and webinars, but I never went to one. I would not have described it as a secret or a society.

–The panels were largely self-organized, so people would propose panels and hold them. I went to one on being a working parent and another on whether crypto had any real use cases and another on how to accelerate scientific breakthroughs. You’d usually have 8 or 10 people in a room. It was all very TED-talk adjacent.

–In 2018, I found it very optimistic, with an idealistic hacker-ish vibe. In 2022, I found the conversations and vibe more curdled and resentful. I didn’t enjoy it, and I didn’t go back. (That did prove a pretty good signal of where tech’s politics were going though, maybe I should’ve paid more attention.)

....

"

first half of his comment about it from X

I was referring to the US government buying a 10% stake in Intel and the ideas being floated of further investment in Intel and other companies by the Trump admin.

It's interesting to see an AI company need to pivot so hard in order to find revenue. I guess this means there is very little easy money to be made as more and more models get created, shared and downloaded by others.

The US is slowly becoming more like China. From talks of nationalizing companies to make US state owned entities to banning foreign competition. It's just so strange how you become the thing you fear.

Wow, thanks for sharing that maps. Looking at the satellite is wild. It's all roads and parking lots, barely any buildings. Even if you did fix the massive road infrastructure, what would they do, give the land back to businesses nearby?

Unpopular take I know. But ads are a source of revenue for much of the free and open internet. The alternatives are paid features that are a regressive tax on poorer people who can't afford them or fork up larger amounts of their discretionary budget.

While popups and bad ad practices have always been a problem, it's sad to see that they became so bad that the response to them is to paygate web content. More and more sites are locked behind paywalls.

I recently switched to the opencode $20 a month plan and am also testing the $5 a month go plan to see if that works. Connected into MiMo or DeepSeek Zen seems to code all day.

I think that this is what OpenAI/Anthropic want but they wont say it publicly. The will be OK with the US banning regulating and banning open source models as it let's Anthropic and OpenAI charge huge premiums to American business clients for their models.

Also the marketing of them getting to say "our models are so dangerous" only a few companies or select users are allowed to use (benchmark) them would help keep their valuations high.

AppGoblin lets any user, from anywhere, request to scan any mobile app for the SDKs, trackers and API calls the app makes.

How it works?

Search the app you're interested in on AppGoblin

Go to the main App page and press the "Scan SDKs & APIs" button

AppGoblin goes and fetches the latest version of the app.

AppGoblin analyzes the Android APK or iOS IPA file for known trackers, ad networks and other business tools that scrape user data.

For Android apps, AppGoblin runs the app in an emulator to track what data leaves the app in the first 60 seconds

The results are prepared and put on the public app page within 24 hrs

I hit the limit and stopped, to my knowledge, using the CoPilot APIs. I switched to open code and DeepSeek which (seems? I'm not an expert in comparing these) cheaper.

Then I went back a couple days later and there was some random $0.xx charges every day from who knows what OpenAI GPT 5.5 calls even though it looked to me like I had my opencode model selected.

Definitely set your overage API to $0 if you are expecting $0 charges after your subscription runs out.

Yes! I am considering cancelling Copilot so was testing OpenCode. I first tried the OpenCode extension and was super disapointed it was just the CLI which I use, but it is nowhere near as good for reviewing changes as the VSCode AI features (ie jumping between files, reading the LLM thought process -> line of code changed etc).

When I realized I could just stick my opencode models in and fully turn off copilot it was a huge win.

I wonder if this will get closed at some point. The LLM models are a cheap commodity, the convenient and power that comes from the direct integration with the IDE is where power users will notice the difference.

Well reported now, but I think China's huge investment in electricity in all forms, Solar, Coal, Nuclear, Hydro will reap huge benefits for citizens and data centers alike.

Additionally, if they continue, which looks likely, to have abundantly cheap electricity, they will be able to out compete high end GPUs by sheer numbers. Cheaper GPUs will also be a commodity that once it gets going will be interesting to see how well they export to consumers / abroad.

I found some 60 iOS apps that have the SDK mentioned in the article: https://appgoblin.info/sdks/brdsdk.framework (sorry this requires a free login due to heavy scraping, feel free to contact me for list)

I was unable to find related Android SDKs. I tried looking at the various apps on AppGoblin to find the android versions, then looking through their unmapped SDK parts but didn't see anything.

https://github.com/BrightSDK/bright-sdk-gradle-plugin-docs

This looks like it should just be "com.brightdata" but I did not find anything. With 60 iOS apps there must be apps with Android SDK, but I'm not sure why I am not finding any.

If anyone knows, or would like to chat feel free to connect. I'm happy to share data.

Most of the time using LLM generated code the feeling is "Oh Awesome!"

My "Uh Oh" feelings are weeks later when I realize there is a subtle bug in what the model presented as test passing "awesome" that I didn't read closely.

The biggest uh-oh is when I get lazy and let it modify multiple files and make many changes at once, and YOLO because I didn't fully understand what it did. I can usually get away with that for frontend, but for data manipulation tasks if I don't understand it, it's likely not what I wanted and I'll be back again in weeks or more trying to figure out what changed.

That's more or less what life was before LLMs and copy pasting from StackOverflow. Most of the time if I didn't fully understand something, I knew I had to eventually get back to it to grok what changed before committing.

Now with LLMs the 'copy pasting' is much faster and handles boilerplate super well letting me focus on edge cases.

What did you use to make the bump chart? I've used eCharts https://echarts.apache.org/examples/en/editor.html?c=bump-ch...

Yours looks cleaner and I love the way you handled the off chart samples with dashed lines, that's a great idea.

Also, using a central area below the chart to show which is highlighted is also great for those that in the past fell off and are no longer on the chart, but naturally people are curious and want to click to see what their names are.

Here's the eCharts BumpChart on AppGoblin, showing the latest Google Play Store rankings: https://appgoblin.info/rankings/store/1/collection/1/categor...

But you can see that doing it that way there it's not really clear where / to the apps that just popped into the top 10 came or went.

I'll try adding the selection area like you have, eg:

Cantique de Jean Racine, Op. 11 (Fauré, Gabriel)

2001: Ranked 100th

2010: No Rank

2021: No Rank

Great little chart, thanks!

Working through the pros and cons of the W3 Attribution framework and how it might impact ad measurement. I also focused a bit on AppsFlyer as it was something I was curious to think about how the browser based Attribution could affect or overlap with AppsFlyer's recent released web2app attribution models.

I just spent an hour reading the WC3 proposal and I indeed think this is the core problem with it, and I came to agree with the original post here, this is a change that will benefit advertising networks and remove transparency for advertisers that want to verify ad networks.

As a user I do not mind when an advertiser (eg Nike) tracks what blog I came from to Nike website. My concern is a Meta / Google that tracks every site I was on and went to. So in this way, I think advertisers and users should be aligned.