It's part of antigravity for free. Just make a blank workspace and ask it to use a browser to do X and it'll start chrome and start navigating, clicking, scrolling, etc.
Yeah, I only found it by accident when I asked it to make a change against my web app and it modified the code then popped open Chrome and started trying different common user/pass combinations to log into the app so it could validate the changes.
Chrome's DevTools MCP has been excellent in my experience for web development and testing. Claude code can jump in there and just pretend to be a user and do just about everything, including reading console output.
I'm not using it for the use case of actually interacting with other people's websites, but for this purpose, it's been fantastic.
The job doesn't pay you to be curious. It pays you to get stuff done. Curiosity makes you jobless. Most of the Silcon Valley people who frequent this website larp as curious people, but are basically incurious status seekers.
Well, you could say mine doesn’t either, literally, but the only reason I am in this role, and the driving force behind my major accomplishments in the last 10 years, has been my curiosity. It led me to do things nobody in my area had the (ability|foolishness) to do, and then it led me to develop enough improvements that things work really well now.
I'd be curious if jobs like yours are not on the tail side of the distribution. It's very common that in work groups, curiosity / creativity gets ignored if not punished. I've seen this even in small techies groups, there was a natural emergence of boundaries in which people don't get to think beyond (you're overstepping, that's not your role, you're doing too much). It seems a pavlovian reflex when leadership doesn't know how to operate without assigning roles.
I mean, think of all the people getting paid eight-digit compensation right now because they were curious about this dead-end deep learning stuff 15 years ago for no good reason!
I couldn't resist... Like the kid at facebook who's buddies with Altman so gets to be a billionare? Like Altman himself (when did he enter the field again? Oh yea he was a crypto huckster). Like everyone I've ever met in the machine learning department? 95% of the people in that field are just following trends and good at winning that game. Call it sour grapes, but I'm just observing reality here. And everyone who thinks following fads = being curious is just doing the larp I described earlier. Moreover, everyone who thinks following fads keeps them safe from AI is deluding themselves. The AI of 2026 can do it better than you can.
Didn't Sam Altman write a "friend locator" app that sold for millions after (angrily) refusing to disclose how many users it had? Then it was summarily shut down after acquisition ... and turned out to have never had more than 500 DAU (though appreantly more registrations)
Fully expecting to get banned for my comment, but I'll just go on. Look at the silicon valley heroes and they're all business types. There's a few rare exceptions.
I don’t necessarily think you’re wrong, but I’m skeptical that the curious will really meaningfully learn from LLMs. There’s a huge gap between reading something and thinking “gee that’s interesting, I’m glad I know that now,” and really doing the work and deeply understanding things.
This is part of what good teaching is about! The most brilliant engaged students will watch a lecture and think “wow nice I understand it now!” and as soon as they try to do the homework they realize there’s all kinds of subtleties they didn’t consider. That’s why pedagogical well crafted assignments are so important, they force students to really learn and guide them along the way.
But of course, all this is difficult and time consuming, while having a “conversation” with a language model is quick and easy. It will even write you flowery compliments about how smart you are every time you ask a follow up question!
I find LLMs useful for quickly building mental models for unfamiliar topics. This means that instead of beating my head against the wall trying to figure out the mental model, I can beat my head against the wall trying to do next steps, like learning the lower level details or the higher level implications. Whatever is lost not having to struggle through figuring out the mental model is easily outweighed by being able to spend that time applying myself elsewhere.
I have some success by trying to explain something to an LLM, having it correct me with its own explanation that isn’t quite right either, correcting it with a revised explanation, round and round until I think I get it.
Sort of the Feynman method but with an LLM rubber duck.
yes and it will mostly depends on the culture / economy. if you create incentives for kids to explore ideas through LLMs they'll become very knowledgeable (and maybe somehow confident). otherwise it will be the tiktok of cognition.
10 bucks there will be a law to enforce exponential backoff so that you need to get good after a few questions before the LLMs delays things by an hour
This is assuming the current AI business model (losing lots of money). As with the internet as a whole, AI companies will probably be incentivized to waste your time and increase "engagement" as they seek revenue. At that point, AI will only be a good tutor if you're extremely diligent at avoiding the engagement bait.
The french product was called Mindie, the start of endless short videos with music, the reels, and the copycat was chinese musical.ly (later bought up by Bytedance).
Mindie failed because they got attacked by copyright claims by the big US tech, which obviously the chinese copycat was immune to, since China.
Your version only describes what happens if you do the operations serially, though. For example, a consumer SSD can do a million (or more) operations in a second not 50K, and you can send a lot more than 7 total packets between CA and the Netherlands in a second, but to do either of those you need to take advantage of parallelism.
If the reciprocal numbers are more intuitive for you you can still say an L1 cache reference takes 1/2,000,000,000 sec. It's "ops/sec" that makes it look like it's a throughput.
An interesting thing about the latency numbers is they mostly don't vary with scale, whereas something like the total throughput with your SSD or the Internet depends on the size of your storage or network setups, respectively. And aggregate CPU throughput varies with core count, for example.
I do think it's still interesting to think about throughputs (and other things like capacities) of a "reference deployment": that can affect architectural things like "can I do this in RAM?", "can I do this on one box?", "what optimizations do I need to fix potential bottlenecks in XYZ?", "is resource X or Y scarcer?" and so on. That was kind of done in "The Datacenter as a Computer" (https://pages.cs.wisc.edu/~shivaram/cs744-readings/dc-comput... and https://books.google.com/books?id=Td51DwAAQBAJ&pg=PA72#v=one... ) with a machine, rack, and cluster as the units. That diagram is about the storage hierarchy and doesn't mention compute, and a lot has improved since 2018, but an expanded table like that is still seems like an interesting tool for engineering a system.
> For example, a consumer SSD can do a million (or more) operations in a second not 50K
The "Read 1MB from SSD" entry translates into a higher throughput (still not as high as you imply, but "SSD" is also a broad category ranging from SATA-connected devices though I think five generations of NVMe now); I assume the "Read 4KB" timing really describes a single, isolated page read which would be rather difficult to parallelize.
Great comment. I like your phrasing "capacities of a reference deployment", this is what I tend to refer to as the performance ceiling. In practical terms, if you're doing synthetic performance measurements in the lab, it's a good idea to try to recreate optimal field conditions so your benchmarks have a proper frame of reference.
Your suggestion confuses latency and throughput. So it isn't correct.
For example, a modern CPU will be able to execute other instructions while waiting for a cache miss, and will also be able to have multiple cache loads in flight at once (especially for caches shared between cores).
Main memory is asynchronous too, so multiple loads might be in flight, per memory channel. Same goes for all the other layers here (multiple SSD transactions in flight at once, multiple network requests, etc)
Approximately everything in modern computers is async at the hardware level, often with multiple units handling the execution of the "thing". All the way from the network and SSD to the ALUs (arithmetic logic unit) in the CPU.
Modern CPUs are pipelined (and have been since the mid to late 90s), so they will be executing one instruction, decoding the next instruction and retiring (writing out the result of) the previous instruction all at once. But real pipelines have way more than the 3 basic stages I just listed. And they can reorder, do things in parallel, etc.
I'm aware of this to an extent. Do you know of any list of what degree of parallelization to expect out of various components? I know this whole napkin-math thing is mostly futile and the answer should mostly be "go test it", but just curious.
I was interviewing recently and was asked about implementing a web crawler and then were discussing bottlenecks (network fetching the pages, writing the content to disk, CPU usage for stuff like parsing the responses) and parallelism, and I wanted to just say "well, i'd test it to figure out what I was bottlenecked on and then iterate on my solution".
Napkin math is how you avoid spending several weeks of your life going down ultimately futile rabbit holes. Yes, it's approximations, often very coarse ones, but done right they do work.
Your question about what degree of parallelization is unfortunately too vague to really answer. SSDs offer some internal parallelism. Need more parallelism / IOPS? You can stick a lot more SSDs on your machine. Need many machines worth of SSDs? Disaggregate them, but now you need to think about your network bandwidth, NICs, cross-machine latency, and fault-tolerance.
The best engineers I've seen are usually excellent at napkin math.
Both ops/sec and sec/op vary on clock rate, and clock rate varies across machines, and along the execution time of your program.
AFAIK, Cycles (a la _rdtsc) is as close as you can get to a stable performance measurement for an operation. You can compare it on chips with different clock rates and architectures, and derive meaningful insight. The same cannot be said for op/sec or sec/op.
Unfortunately, what you'll find if you dig into this is that cycles/op isn't as meaningful as you might imagine.
Most modern CPUs are out of order executors. That means that while a floating point operation might take 4 cycles to complete, if you put a bunch of other instructions around it like adds, divides, and multiplies, those will all finish at roughly the same time.
That makes it somewhat hard to reason about exactly how long any given set of operations will be. A FloatMul could take 4 cycles on it's own, and if you have
FloatMul
ADD
MUL
DIV
That can also take 4 cycles to finish. It's simply not as simple as saying "Let's add up the cycles for these 4 ops to get the total cycle count".
Realistically, what you'll actually be waiting on is cache and main memory. This fact is so reliable that it underpins SMT. It's why most modern CPUs will do that in some form.
I agree that what you're saying is true, but in the context of my comment, I stand by the statement that cycles/op is still a more meaningful measurement of performance than seconds.
---
Counter-nitpick .. your statement of "if you put a bunch of other instructions around it" assumes there are no data dependencies between instructions.
In the example you gave:
FloatMul
ADD
MUL
DIV
Sure .. if all of those are operating on independent data sources, they could conceivably retire on the same cycle, but in the context of the article (approximating the performance profile of a series of operations) we're assuming they have data dependencies on one another, and are going to be executed serially.
Your critique applies to measuring one or a handful of instructions.
In practice you count the number of cycles over million or billion instructions. CPI is very meaningful and it is the main throughput performance metric for CPU core architects.
In cpu uarch design, sure, but that's outside the context of the discussion. There's nothing you can do to that C++ library you are optimizing that will impact performance due to register allocation/renaming.
This is not always true. Compilers are quite good at register allocation but sometimes they get it wrong and sometimes you can make small changes to code that improve register allocation and thus performance.
Usually the problem is an unfortunately placed spill, so the operation is actually l1d$ traffic, but still.
The reason why that formatting is not used is because it’s not useful nor true. The table in the article is far more relevant to the person optimizing things. How many of those I can hypothetically execute per second is a data point for the marketing team. Everyone else is beholden to real world data sets and data reads and fetches that are widely distributed in terms of timing.
Although cool, I can see how this product will just inhibit instead of enabling creativity and play in kids. Instead of having to draw something to see it, refining the drawing over minutes or hours, the kid will just lazily ask for some half formed idea, and see it materialize from thin air. That's just sad
I think with the right parental guidance/supervision this could be a very fun toy.
From the website it seems like a great way to generate some black and white outlines that kids can still color in. If used like that it seems almost strictly more creative than a coloring book, no? There are plenty of other ways kids can express creativity with pre-made art too. Maybe they use them to illustrate a story they dreamed up? Maybe they decorate something they built with them?
Also, some children might want to have fun be creative in ways that don't involve visual arts. I was never particularly interested in coloring or drawing and still believe myself to be a pretty creative individual. I don't think my parents buying me some stickers robbed me of any critical experience.
Yeah. It's bad enough if kids prompting this stuff online is the new form that creativity is going to take. But this way, it's generating electronic crap that will end up in landfills as well.
The amount of permutations of words you can say along with the permutations of drawings you can add to the sticker is a gigantic number. That is a big enough state space for people to express creativity.
The entire point of being creative is that you actually MAKE it yourself, not that you tell the slop machine to make it for you. This is, quite frankly, the complete and total opposite of creativity. This is pure consumption disguised as creativity, wrapped up in a nice $99 box that will probably be e-waste in a couple years when the company goes under.
Is writing a book not creative if you use an existing font? There are both creative aspects like choosing words or choosing a font and non creative aspects like rendering the font.
You and I both know this is a disingenuous argument. You're not saying "Hey, Stickerbox. Choose a nice font for me." You're saying: "Hey, Stickerbox, write the entire book for me."
Google allows AI browser automation through Gemini CLI as well, but it's not interactive and doesn't have ready access to the main browser profile.
reply