ericskiff

Thank you, this is a huge pet peeve of mine. I mis-click my airpod and suddenly this app I've never used and don't want launches

Just chiming in for anyone else reading this and worried about their hair thinning, this is a thing men worry about much more than women. Bruce Willis, Patrick Stewart, Michael Jordan, Vin Diesel - there have been plenty of bald sex symbols and plenty of women who enjoy that look (and the look of many average men of various builds and ages)

The factors of attractiveness are by far more related to basic self care (hygine and being fit enough to care for yourself and others), kindness and the ability to share in others' joy, and passionate interest in something, and a lightness or humor in your manner.

I started balding at 18, shaved it all at 22. Its not an issue. Height, hair, etc might get immediate reaction and attention, but hardly matter for real connections.

What's interesting is that in this article, the author describes making an understandable mistake (accidentally deleting Trunk aka main from source) and how their team was able to easily recover from that due to the nature of SVN.

The actual "AI deleted my database" story is really more of a "Railways' database 'backup' strategy is insane and opaque and Railway promoting AI infrastructure orchestration without guardrails is dangerous."

If removing Trunk had irrevocably deleted it from a single centralized server and also deleted any backups of it, there would have been an "SVN and the CLI destroyed our company" article back then.

As a Railway user, I appreciated that information and have changed my strategy when using them.

That comment seemed to revolve around consent. Willful, nonconsensual dosing of anyone with any drug is a violation, and yes doing it and bragging about it is reprehensible.

Fun, but the way they fly doesn't quite match my intuition. Why would an object curve when I send it out on the tangent? Wouldn't that be a straight line unless it's affected by a different gravity well?

Wooooow, I LOVED this site when it first came out and I still reference it when I talk about the early web and how it enabled me to learn things I never would have otherwise. I can still picture the animation of the wankel rotary engine from this site whenever I think about it.

This and howstuffworks.com made me so hopeful for the future of the web when I was young

This is so lovely! If the original author is here in the comments, some feature requests that would absolutely make my day, presumably from easiest to hardest :)

I love this so much, thank you for sharing!

* Slow down the motion to about .5 of what it is currently, with easing/acceleration on the speed to emulate the camera dolly and jib effects used in the film

* Add a random motion setting that allows me to run it full screen just sliding through the aisles, banking around turns, flying up and then back down the aisles.

* optionally lock the framerate to 24fps to give it a film feel

* optional shaders on the main viewport to emulate lens distortion, film grain, etc

* raytracing with reflectivity on the glass, refraction, diffusion, etc.

For anyone taking this comment seriously, please research and understand the potential long term impacts of GBL before going near it. It's neurotoxic and can cause brainfog and lowered cognitive ability. It's also lethal in the wrong dose, with a tiny margin for error.

It's by no means a safe alcohol replacement

Awesome to see real UX experimentation, and this elicited a strong response from me at first "oh I haaaaate that".

On further reflection, this is very interesting and I understand where the drag and drop interaction breaks down on long lists. Some additional UI affordances to communicate what's happening may make it intuitive and clear.

Things I'd want to experiment with if I was implementing this:

* A "wheel" effect where the items in the list grow slightly as they near the chosen item which stays locked in the interface at the center, popping into place at at each 'click'. Somewhat like the Price Is Right wheel flipper

* Making the interaction entirely scroll based once I click. Setting the item in place can be done by any other click or keypress, and cancelled with the escape hotkey. My interaction is pick, scroll, click (without having to aim back at the thing I just placed by scrolling)

I believe this was a statement on cost per token to us as consumers of the service

It all reads like hallucinated slop from top to bottom

"I've been tracking software quality metrics for three years" and then doesn't show any of the receipts, and simply lists anecdotal issues. I don't trust a single fact from this article.

My own anecdote: barely capable developers churning out webapps built on PHP and a poor understanding of Wordpress and jQuery were the norm in 2005. There's been an industry trend towards caring about the craft and writing decent code.

Most projects, even the messy ones I inherit from other teams today have Git, CI/CD, at least some tests, and a sane hosting infrastructure. They're also mosty built on decent platforms like Rails/Django/Next etc that impose some conventional structure. 20 years ago most of them were "SSH into the box and try not to break anything"

go to tax.gov

You'll identify on id.me

People have just gotten used to this sort of thing unfortunately

Folks have been predicting that the next big shift in computing will be onto glasses that we wear and away from our phones.

The tech just hasn’t been there yet and most of the devices that do this are heavy clunky and hot

Meta is investing billions to get out ahead of this shift and to own the entertainment and data (and thus advertising) layers that sit on top of the real world through these glasses

The rumor mill is abuzz that Facebook finally making a play for it in the next set of smart glasses after a few years of sticking to VR headsets and audio/camera only glasses

Has anyone gotten access to Imagen 4 for image editing, inpaint/outpaint or using reference images yet? That's core to my workflow and their docs just lead to a google form. I've submitted but it feels like it's a bit of a black hole.

I highly recommend the command line AI coding tool, AIder. You fill its context window with a few relevant files, ask questions, and then set it to code mode and it starts making commits. It’s all git, so you can back anything out, see the history, etc.

It’s remarkable, and I agree Claude 3.5 makes playing with local LLMs seem silly in comparison. Claude is useful for generating real work.

What we can reasonably assume from statements made by insiders:

They want a 10x improvement from scaling and a 10x improvement from data and algorithmic changes

The sources of public data are essentially tapped

Algorithmic changes will be an unknown to us until they release, but from published research this remains a steady source of improvement

Scaling seems to stall if data is limited

So with all of that taken together, the logical step is to figure out how to turn compute into better data to train on. Enter strawberry / o1, and now o3

They can throw money, time, and compute at thinking about and then generating better training data. If the belief is that N billion new tokens of high quality training data will unlock the leap in capabilities they’re looking for, then it makes sense to delay the training until that dataset is ready

With o3 now public knowledge, imagine how long it’s been churning out new thinking at expert level across every field. OpenAI’s next moat may be the best synthetic training set ever.

At this point I would guess we get 4.5 with a subset of this - some scale improvement, the algorithmic pickups since 4 was trained, and a cleaned and improved core data set but without risking leakage of the superior dataset

When 5 launches, we get to see what a fully scaled version looks like with training data that outstrips average humans in almost every problem space

Then the next o-model gets to start with that as a base and reason? Its likely to be remarkable

We’ve been hitting this in our work and in experimentation, and I can confirm that Claude sonnet 3.5 has gotten 100% of the way there, including working through errors and tricky problems as we tested the apps it built.

Interestingly, I’m pretty sure they mean they hit the limit with tokens on Claude.

There’s a daily 2.5 million token limit that you can use up fairly quickly with 100K context

So they may very well have completed the whole program with Claude. It’s just the machine literally stopped and the human had to do the final grunt work.

Deskpad might be what you’re after! It’s a virtual display in a window, you can share that instead of your whole screen but still get multi-app flows captured

https://github.com/Stengo/DeskPad

100% Being able to write, code, and read from a large virtual terminal without a clunky visor on would be amazing.

I’ll share my experience and the experience of my kids so far.

Aside from blindly copying and pasting a response, in which case the learner wasn’t interested in learning and probably would have plagiarized from somewhere else anyway, I have found LLM to be an incredible, endlessly patient teacher that I’m never afraid to ask a question of.

My kids who are in the tween and teenage years, are incredibly skeptical and dismissive of AI. They regard AI art as taking away creative initiative from artists and treat LLM similar to the way we treated Google growing up, if they use them at all. It’s a tool which can be helpful for answering questions that is part of the landscape of their knowledge building.

That knowledge acquisition includes school, YouTube and other short videos, their peers (online and off) Internet searches, and asking AI. Generally, I regard asking AI as one of the least problematic sources of info in that environment.

While I tend to be optimistic as a default, I truly do think that the ability to become less ignorant by asking questions is a net positive for humanity.

The only thing I truly lean on AI for right now is as an editor, helping me turn my detailed bullet points into decently crafted prose, and for generating clear and concise transcripts and takeaways from long meetings. To me that doesn’t seem like the downfall of human knowledge.

For over 10 years we've had people do a paid ~4 hr take-home which is very similar to the work they'll actually be doing (here's a dummy codebase, add a few features fix a few bugs).

If they're not interested in getting paid to do that work now, it's a good signal for us that they won't be happy doing it when they're working with us. It's helped us find really wonderful people to work with.

Counterpoint - I pay for my whole team to have access, shared tools, etc. we also spend a decent amount on their APIs across a number of client projects.

OpenAI has a strong revenue model based on paid use

I've been playing with Continue: https://github.com/continuedev/continue

Thinking about how this may apply to AI, this research seems to be saying that the distribution of neurons is not a normal distribution, but more akin to the population of a country, where there are dense clusters of neurons amid a mostly sparse network, the same way populations heavily cluster in urban areas.

We don’t yet know the reason for that structure, but it appears across species and across different regions of the brain

It could be that a similarly modeled neuronal network would exhibit more pockets of specialized activity in response to certain inputs, and potentially even redundancies in specialization allowing it to process the inputs in slightly different ways to reach ensemble/consensus style outcomes. In some ways it sounds reminiscent of the mixture of experts approach

It’s certainly interesting!

So this study seems to be about maximizing the nutrients from sea catches to consumption, suggesting we should eat what we feed the salmon instead of the farmed salmon themselves, but anchovies, herring, and mackerel aren’t on most people’s menu often.

This study is not about whether farmed or wild caught fish are better for you.

Personally, it’s never been about free shipping for me. It’s shipping inclusive pricing.

It’s the same reason I will always toggle the “include fees” on Airbnb’s search interface. Just tell me the real price.

Upfront pricing and an assumption I’ll get the item in 1-3 days is the real value prop, and I’m still a happy customer of that product

We seem to be at an interesting feedback loop point for ML models.

We have textual and visual models strong enough to look at a picture and describe it, or to read a body of text and categorize and label it well.

That work can then be used to feed the training of other models, whether it's larger models that can make use of an ever-growing well labelled corpus, or smaller and more efficient models trained on sets that were previously too niche or cost prohibitive to individually label.

With this coming out of Microsoft for copilot/Bing and the strength of recent smaller models which have had the benefit of GPT4 and other larger LLMs to assist with training, we appear to be at an inflection point of training quality at the same time as training compute is being massively scaled.

Very interesting! I was curious how this related to Roman self-healing concrete: https://news.mit.edu/2023/roman-concrete-durability-lime-cas... and it seems like this is an entirely novel method, where little capsules of bacteria break open when a crack appears and then "heal" the crack through growth until they consume their food stock.

Both the capsule (alginate-based) and bacteria are common and safe

Has anyone found the context length for these models yet? So far I haven't seen it mentioned in their write-up or the model card