felipeerias

Are you sure about that? Every information I’ve seen suggests that the DoD has been using Anthropic’s models through Palantir.

My understanding is that Anthropic requested visibility and a say into how their models were being used for classified tasks, while the DoD wanted to expand the scope of those tasks into areas that Anthropic found objectionable. Both of those proposals were unacceptable for the other side.

If it turns out that there is significant value in everyone having their own personal agent running 24/7, we might end up needing a lot more compute than anticipated.

(It’s a big if! I’m not convinced about that myself, but it’s worth considering that possibility.)

Civilisations in the Americas were significantly less technologically developed than those in Eurasia. We focus our analysis on the Spanish and Portuguese, but the outcome would not have been much different had their place been taken by the Ottoman or the Chinese.

The Mayan and the Aztecs were roughly at a similar level of development as ancient Sumer or Babylon: good agricultural practices, irrigation, astronomy, elaborated culture, rich mythologies, very basic metallurgy, early state structures, etc.

Sumer and Babylon were great civilisations whose legacy can still be traced today. The same is true for the Maya and the Aztec. Had you visited any of them in their prime, you would have been awed by their skill and sophistication.

And yet, think of everything that happened in Eurasia between Hammurabi and Columbus, and you will get a sense of how wide the gap was when the two worlds met.

"Here's the thing nobody tells you", "here's the part that should unsettle you"…

I am European and live in Japan.

China is currently providing weapons for Russia to wage large-scale war in Europe, while supporting a dictatorship in North Korea that regularly launches nuclear-capable misiles in my general direction.

If you were looking for a good long-term AI benchmark, “build me a Web browser” should last you for a while.

Does that automatically translate into more openings for the people whose full time job is providing that thing? I’m not sure that it does.

Historically, it would seem that often lowering the amount of people needed to produce a good is precisely what makes it cheaper.

So it’s not hard to imagine a world where AI tools make expert software developers significantly more productive while enabling other workers to use their own little programs and automations on their own jobs.

In such a world, the number of “lines of code” being used would be much greater that today.

But it is not clear to me that the amount of people working full time as “software developers“ would be larger as well.

If AI tools make expert developers a lot more productive on large software projects, while empowering non-developers to create their own little programs and automations, I am not sure how that would increase the number of people with “software developer” as their full-time job.

As discussed elsewhere, it is apparently possible to compile and run this particular project. It seems that whatever process they followed allows commits to break the build pretty often.

Nevertheless, IMHO what’s interesting about this is not the browser itself but rather that AI companies (not just Cursor) are building systems where humans can be out of the loop for days or weeks.

IMHO people are missing the forest for the trees. The point of this experiment is not to build a functional browser but to develop ways to make agents create large codebases from scratch over a very long time span. A Web browser is just a convenient target because there are lots of documentation, specs and tests available.

Claims based on personal experience working on real world problems are likelier to be true.

It’s reasonable to accept that AI tools work well for some people and not for others.

There are many ways to integrate these tools and their capabilities vary wildly depending on the kind of task and project.

The adoption of AI tools for software development will probably not result in sudden layoffs but rather on harder to measure changes, like smaller teams being able to tackle significantly more ambitious projects than before.

I suspect that another kind of impact is already happening in organisations where AI adoption is uneven: suddenly some employees appear to be having a lot more leisure time while apparently keeping the same productivity as before.

On many fields, the link between “writing papers” and “producing science” was already fraying before the arrival of LLMs.

We will have to find better ways to share and promote valuable research, before we all drown in the noise.

Yes, perhaps the most reasonable approach is to tie these restrictions to the specific device (and, where applicable, account).

The ban doesn’t need to catch every single case, it just needs to add enough friction to stop the most frequent and destroy network effects.

Some CEOs are obviously extremely skilled at marshalling large organisations in the successful pursue of business goals, while the rest derive most of their value from sharing a job title with the first group.

Things seem to be heading in the direction of using formal languages to define deterministic behaviour and natural languages to express matters of human taste.

The most important point is that even if AI capabilities do not improve much past the current state of the art, the consequences of this technology will still take years to play out, both in tech and in society as a whole.

We are already seeing diminishing returns on state of the art models as training becomes more expensive for smaller relative improvements. For AI companies, attempting the next generational leap becomes an increasingly risky bet, but missing the next generational leap would mean bankruptcy.

In the rest of the tech world, attention and investment will continue to move towards tools build on top of AI models, probably with coding agents as the initial and main focus. AI is a foundational technology, but we are still figuring out what we will build on that foundation.

This misunderstands what LLM-based tools mean for complex software projects. Nobody expects that you should be able to ask them to write you a whole kernel or a web engine.

Coding agents in particular can be very helpful for senior engineers as a way to carry out investigations, double-check assumptions, or automate the creation of some parts of the code.

One key point is to use their initial output as a draft, as a starting point that still needs to be checked and iterated, often through pair programming with the same tool.

The mid-term impact of this transition is hard to anticipate. We will probably get a wide range of cases, from hyper-productive small teams displacing larger but slower ones, to AI-enhanced developers in organisations with uneven adoption quietly enjoying a lot more free time while keeping the same productivity as before.

As the article points out, the more of the original color scheme that has survived, the better the reconstructions look.

Example: https://www.dailyartmagazine.com/wp-content/uploads/2022/09/...

The author suggests that this minimizes the opportunity for mischief, but tbqh it's likely that the ancients were simply much better artists than the people carrying out these reconstructions today.

I'd love to see a modern artist attempt one of these reconstructions using original materials but with greater artistic freedom.

Thinking about the relationship between creation and verification is a good way to develop productive workflows with AI tools.

One that works particularly well in my case is test-driven development followed by pair programming:

• “given this spec/context/goal/… make test XYZ pass”

• “now that we have a draft solution, is it in the right component? is it efficient? well documented? any corner cases?…”

Tools can make individuals and teams more effective. This is just as true for LLM-based tools as it was for traditional ones.

The question is not whether one (1) LLM can replace one (1) expert.

Rather, it is how much farther an expert can get through better tooling. In my experience, it can be pretty far indeed.

In this context, time constraints are measured in hours and are very informative regarding the student’s capacity to prioritise, plan and carry out their work under pressure.

It is actually very informative when one person can

In these studies, the qualitative data is often a lot more informative than the quantitative.

Understanding how concrete people navigate a domain and noting the common points between them can be illuminating.

Trying to calculate a generalisable statistical result from them… probably not so much.

I share the sentiment but it seems a bit like imposing a human narrative on a Universe that does not seem to care all that much about us. Maybe we really are stuck in the Solar System and space is just too vast to do much about it.

In your example, you could achieve a similar outcome with a skill that included a custom command-line tool and a brief description of how to use it.

MCPs are specially well suited for cases that need a permanent instance running alongside the coding agent, for example to handle authentication or some long-lived service that is too cumbersome to launch every time the tool is called.

If I am not mistaken, Bump still required a connection to the Internet. WiFi Aware does not, because the phones create an ad-hoc link on the spot.

The connection can be very fast. In this example, a 280 MB file is transferred in less than 10 seconds:

https://vimeo.com/418946837

I was experimenting with this technology almost a decade ago as part of my work as interaction designer:

https://darker.ink/writings/Mobile-design-with-device-to-dev...

It has a lot of potential but unfortunately it has been kept back until now by lack of support and interoperability.

IMHO coding use cases are much more constrained by tooling than by raw model capabilities at the moment. Perhaps we have finally reached the time of diminishing returns and that will remain the case going forward.

It seems very naive to presume that a tool which explicitly works by unblocking the retrieval of harmful information will not be used for, among other purposes, retrieving that same harmful information.

Users will ask ChatGPT for recommendations and the answer will feature products and services that have paid to be there, probably with some sort of attribution mechanism so OpenAI can get paid extra if the user ends up completing the purchase.