patrickhogan1

Obsidian is amazing

Common Crawl, while a massive dataset of the web does not represent the entirety of the web.

It’s smaller than Google’s index and Google does not represent the entirety of the web either.

For LLM training purposes this may or may not matter, since it does have a large amount of the web. It’s hard to prove scientifically whether the additional data would train a better model, because no one (afaik) not Google not common crawl not Facebook not Internet Archive have a copy that holds the entirety of the currently accessible web (let alone dead links). I’m often surprised using GoogleFu at how many pages I know exist even with famous authors that just don’t appear in googles index, common crawl or IA.

Thank you! Great example!

Wow. This matches my experience pretty closely. Haven't put GPT-5 Pro through its paces much yet, but these numbers suggest I should.

That’s a great explanation.

YC genuinely changed the game by making seed funding more accessible. Paul Graham, Jessica Livingston, and the entire YC team saw how broken the system was and fixed it.

It was ridiculously hard to raise seed before YC. I remember pitching a group of 8 angry partners straight out of a cigar filled banker like black and white movie around 2013.

Credit where it’s due. They fixed a broken market.

Do you believe that creating sub agents is a violation of the bitter lesson or is simply a way to add more context?

Do you have a concrete example of what you mean?

For example, the article above was insightful. But the authors pointing to 1,000s of disparate workflows that could be solved with the right context, without actually providing 1 concrete example of how he accomplishes this makes the post weaker.

Keyboard small. Fingers big.

I agree here but also believe it was a way to expose better models to the masses. o3 was so spectacularly good. But a lot of people were still not using it. Even some of my friends who use ChatGPT daily I would say are you using o3 and get a blank stare.

So I think it’s also a way to push reasoning models to the masses. Which increases OpenAI’s cost.

But due to the routing layer definitely cost cutting for power users (most of HN)… except power users can learn to force it to use the reasoning model.

Except the reasoning model o3 and GPT5 thinking can get the right answer. Humans use reasoning.

That because you don’t say

“Think hard about this” and the OpenAI router layer routed you to the cheaper model.

GPT5 seems to violate Rich Sutton’s bitter lesson. As GPT5 makes a lot of human knowledge assumptions about whether to send your prompt to the cheap model or to the smarter more expensive model.

I’m being down voted. I don’t have an agenda. I’m simply sharing my experience. If you’re getting good results with Gemini CLI as an alternative to Claude Code, please let me know what you’re doing to get that performance.

I’m impressed by Gemini Pro 2.5’s NLP capabilities. I use that model in production on several projects. My comments are directed only at Gemini CLI. Which FWIW is better than OpenAI Codex CLI, but much worse (for me) than Claude Code.

Even with Pro, the strict token limits combined with the model's tendency to add unrequested modifications means I run out of tokens before completing my intended tasks. Others have the same issue https://github.com/google-gemini/gemini-cli/issues/4300

I agree with you at this point. Even though Google is performing well on benchmarks and releasing impressive models like World Models Genie 3, the Gemini CLI suggestions/changes feel overly formulaic. Almost like its priorities are that of an OCD coder that cares more about tabs vs spaces instead of building a useful feature. For example, in a recent project, Google CLI spent all of my token allotment for that day on trivial tasks like tweaking ESLint configs or modularizing code that didn't need modularization.

In contrast, Claude Code seems to interpret my prompts better and helps me ship real product features for users.

Maybe it’s a system prompt issue. Its likely my prompting causing the problem. But Claude Code seems to understand my intent better.

Intel N100 devkits are primarily made in Oregon and Arizona in the US.

Realtek devkits are made in Taiwan.

China is clearly important. Nothing here changes that. Orders will just shift to bulk and get sourced from local distributors.

Most “mission-critical” R&D parts will still clear in < 48 hours. Express carriers already transmit full data to CBP before wheels-up. Type 86 filings have been required “upon or prior to arrival” since Feb 2024, so the paperwork is literally done while the flight is in the air.

This will shift important inventory to local distributors in the US. This makes local supply chains more resilient, not more vulnerable.

The US is eliminating its $800 duty-free import threshold. This isn't radical. It mostly brings America in line with what other major markets already do (or are about to do).

EU: Only packages under €150 ($160) are duty-free, and they're planning to scrap even that. Since 2021, they collect VAT on everything, no matter how small.

Japan: Has a ¥10,000 ($70) threshold now, but already announced they're killing it by late 2026.

China: Basically no threshold at all. You only avoid duty if the calculated tax is under $7. Even their special e-commerce imports pay 70% of normal VAT.

For a typical Amazon FBA seller importing bulk orders (eg 1,000+ items), who orders and stores them in an Amazon warehouse AFAIK eliminating de minimis doesn't affect their imports much since they already pay duties.

Direct-to-consumer overseas companies (Temu/Shein style) yes this is a big change.

Is having more tinkerers or Bill Nye's really a bad thing?

From what I’ve seen at maker and science fairs, these events often attract students who feel overlooked in schools that heavily prioritize sports. How many schools have pristine football fields, while the physics teacher is spending money out of their own pocket to build hands-on experiment kits, just to show students that physics is more than what’s in a textbook? (That was the case for my dad)

These fairs open kids eyes to a broader world. One that celebrates creativity, problem-solving, and scientific curiosity.

Not every student needs to become an engineer. What matters is that they feel hopeful about the future and engaged in something positive, instead of turning to drugs or escapism.

Yes the company that has the least competitive, least desired chips, must either raise its prices or cut costs by reducing support to get higher margins! That’s the real issue!! /S

Throw the financial people out. Get the builders and chip designers back in control. Andy Grove is rolling in his grave. FFS.

On your intelligence graph where it shows Grok 4 and OpenAI o4-mini as comparable (and among the highest intelligence rated models), it doesn’t have OpenAI o3 or o3-pro.

Yet all of my tests show o3 blows o4-mini out of the water.

What are you classifying as intelligence?

Plaque buildup is highly heritable. So is the initial biome and often eating patterns since your parent feeds you.

I completely agree, so many people miss out on great experiences simply because they’re afraid of looking dumb.

One big advantage many athletes had growing up was a parent who taught them a skill [basketball, soccer, biking] early, someone they felt safe failing in front of.

That’s one of the things I find so cool about learning with AI, you get to try, mess up, and improve without judgment.

Awesome! This is great!

The link in the article to the full blog explaining rerankers is 404ing for me.

Questions to you as an expert related to search ranking. With o3 and source quality thresholds when performing web search. Could we implement an ELO-style cutoff where systems default to “I don’t know” rather than citing low-ranked sources?

Currently o3’s main weakness is mixing high-quality sources with poor ones when it uses the web search in the same response. The answer sounds authoritative throughout, but parts are backed by unreliable sources. This makes it harder to trust even the well-sourced portions (e.g. believing the US election is next year - not a hallucination but a poorly date formatted source it used). It also makes the response a lot slower.

Would a hard quality threshold be better than the current approach of seamlessly blending good and bad sources?

We need to move past these debates or we’ll slip further behind. China is repeating Japan’s trajectory: early products may have been rough, but Toyota grew into an innovation powerhouse. No one questions the Prius’s impact today.

At some point, we need to stop talking about what we did in high school 10 years ago and start figuring out how to make everything we interact with better by tapping into the collective intelligence of 8 billion people.

What country and what models were they selling?

A UBS teardown estimated the BYD Seal’s (premium version) gross margin at roughly 16%. These cars are selling for $50k in Thailand. These data points suggest BYD is leveraging vertical integration, not selling the Seal at a loss.

I used to think the same way until this year on a trip to Asia. I was genuinely surprised by the range of new Chinese cars. Not just the low-cost models, but truly impressive, high-end vehicles with super innovative designs. I am very pro local manufacturing. But as a builder at heart, I have to admire the builder innovation energy coming out of Chinese auto companies. This wasn't just centrally planned... Its clear innovation.

This really inspired me. I think I will do one on basketball.

Great work on data collection for 10 years. Quantified self tracking into a universal format is still really hard.

What tracking sensors do you use that input into Strava? (e.g. Garmin, Apple Watch, built in iphone, etc)?

This requires audio to work and appears to create more transparency. You can request the audio recording to verify accuracy. This will happen as a routine procedure from defense attorneys. Any problems with the technology would be discovered quickly and if the officer didn’t do their job of correcting the errors before the report is generated they would be torn apart.

Each participant wore a wrist light tracker for 1 week.

Disagree. All work including interesting work involves drudgery. AI helps automate drudgery.