honorious

CaMel paper was posted several months ago. Has anyone experimented with it in concrete examples? Any success stories?

This demo shows an Agent Development Kit (ADK) implementation that leverages the CaMeL framework for enhanced security and controlled data flow in LLM agents. CaMeL (Defeating Prompt Injections by Design) protects the model against prompt injection attacks by explicitly separating control and data flows in the query given to the agent. Additionally, CaMeL enables fine-grained access control; in other words, it is possible to define precise rules that are deterministically enforced over data flows between tool calls.

Am I the only one that thinks that being "agent wrangler" actually makes building things more fun?

To me, the interesting parts in building is taking a real problem, mapping it to a set of "things" that need to be built, decompose them into treatable chunks, and defining how they should interact. This is where intelligence comes in. if agents want to take the rest, please do! I can focus on making better products.

The law says that US cloud providers are fined if they continued to provide services to Bytedance.

As far as we know, Tiktok is operated on US servers by Oracle. While it might have been possible to find another cloud provider and move all US data there, I can see them not wanting to do that given that there was no point if the app isn't distributed in the US anymore.

Can you expand on the risk of breaking novelty?

Is the concern that prompts could be re-used for training by the provider and such knowledge become part of the model?

I have been interested in Prolog since my time at the University, and I loved the idea of logic programming.

For "proper" Prolog, in 2024 it is a niche language alive in specific constraint solving applications, but not really used outside of that. I haven't seen anyone attempting at using prolog as a general purpose language since the 90'.

Datalog and logic-inspired languages tend to pop up here and there as domain-specific languages.

Rego is a recent incarnation which had good adoption for k8s and other "modern" systems. However, when trying to get people in my org to adopt it in practice, I saw engineers struggle with the paradigm when complexity grows to more than toy problems.

I find very appropriate that a Pope, who holds the modern title of "Pontifex Maximus", was the one that finally updated the calendar again after 1500 years.

No need to doubt his story. Very likely he took a boat, like many touring cyclist doing the Pan-American route do. It is a relatively established route.

If you really want to be self-reliant, the amazing Iohan Gueorguiev bike-raft it, but that's quite on the extreme side.

I got out of academia, but I'd love to keep reading the new hot and influential papers to keep up with the research trends in the research communities I am interested in. So I like this idea!

However, there are so many junk papers that the paper title itself is rarely useful. Too much context is lacking: where was it published? How strong is the conference / journal? When was it published? How many citations so far? What sub-sub-sub-field is it in?

Isn't RSU the middle ground already? You can part of your compensation in salary, part in ownership of the company. If you work and the company does well, you do better.

The % of comp in RSU raises as you get promoted to higher levels, where you (theoretically at least) actually have more impact on the success of the company.

What do you like to do outside of work? Do you like to be be near mountains? Near water? Do you like a bars and concerts? Do you like cycling? Climbing? Is there a community for your hobby in the place where you are going to go?

Those activities are also the way you make friends and expand your social circle, which is going to be a big hard thing to do after the move.

As other they are saying, take mini month-long vacation to try out different places, given that you have the opportunity of doing that.

As everything carbon-fiber, a lot depends on the quality of the build.

Look for "Danny MacAskill Tests Santa Cruz Reserve Carbon Wheels": he really tries to destroy those rims, and it takes him doing ridiculous things to do so. Aluminum rims would have cracked/ bent much earlier.

It might not be clear from the last chapter, but the book doesn't claim that time doesn't exist.

To paraphrase (but use the same example because it is funny): time doesn't appear at the fundamental concepts in physics. Cats also don't appear at the fundamental level of physics. But cats exist. Time (and our perception of it) must appear somewhere as a emergent feature.

I just finished reading Rovelli's book "The Order of Time." The article is the last chapter, where the author summarizes the results of the "journey" of the book.

I recommend the book, it was an enjoyable read. The author is able to explain clearly concepts that I had struggled to understand before.

Throughout the book there are a many references to philosophical ideas and to the history of science, in some cases quite emotional. This being the last chapter, is quite heavy on them. However, I found them quite useful and interesting as a tool to break from preconceived notions of time.

Thanks! The design tradeoff between ridership and coverage is very interesting, and ripe for political discussion.

I wonder if there are opportunities for working with on-demand ride companies for coverage, while focusing infrastructure on high-ridership corridors. Fairness of access for disadvantaged people would be a problem to solve.

That looks like an interesting problem!

1) What are the metrics that mass transit operators tend to optimize for? Do you think they are in line with improving the service to customers?

2) How much is "dependability" of a mass transit service a metric that operators are interested in?

Intuitively I feel that there are places (e.g. Chicago) where I know that I can get out of the house, get the bus/train, and get to my destination with little variance in arrival time.

Others (like SF), where the variance on the arrival time seems much higher: e.g., generally related to unexpected delays while waiting for train/bus.

Do you think it's possible to optimize on such factors?

I would argue that abuse protection has been one of the main drivers toward centralized systems.

Many goals of fully decentralized morality-neutral platforms are in conflict with the expectations of people to be protected from harm while using the internet. Centralized platforms get closed to providing such a protection (even if flawed).

There is tons of research in decentralized methods for protecting against abuse (huge research topic in the 2000), yet not much has really worked (Bitcoin solves SOME, but it still suffers of many other abuse problems).

Small-scale federated solutions (where all participants know each other) are potentially better on that front, but I don't think they will be profoundly different from solutions where control is centralized. There will need to be coordination in response to problems, hard to see how different parts of the system might implement profoundly different policies about content, anonymity without splitting the group.

So, my conclusion is that solutions with centralized control are dominating the market not because a few evil corporations are conspiring to steal the control away from the free people, but because they DO provide a much safer environment for regular people to do regular things, despite all downside.

I have reviewed and published several papers for conferences in Computer Science. You got good suggestions so far, I'd like to add one important thing:

* Choose and Know your community!

When you choose a venue like a conference, you are writing an article with the intent of being read by a very specific subset of the research community. You can imagine each subgroup having a set of "interesting conversations" around a narrow set of topics or methodology that are deemed important by the community.

So ... you have to convince your audience that you are contributing something to their conversation. Go look up papers in the previous edition of the conference, dig into their references, and frame your contribution in the term of the conversation they are having. In this way it'll be easier for everybody reading the paper to understand what you are doing.

In some cases, it might be that you are telling the community that they should care about this new problem of yours, but if it is completely unrelated to their discussion, it'll be an harder sell.

Also, be aware of the style of the community: are they interested in experiments backed by strong theoretical work? Or are they more interested in practical experiences, without caring much about theory? A decent paper accepted in a theoretical conference might actually be rejected by a more practical conference (and the other way around).

It's safe to assume that most US universities with PhD programs have openings every year and they are going to sponsor you for a visa.

The difficult part is identifying professors that you'd like to work with.

PhD students take classes, and some of them do have written exam (in CS theory more than other parts of CS). In the US, where the PhD is 5 years, you spent a good portion of the first 2 years taking classes.

Before coming up with new knowledge you need to know what's already out there.

One thing that comes to mind is G+ pictures: it has the sharing functionalities you are planning to have, and it has a lot of more features in term of picture processing.

How do you compete? I can see a few things (maybe simpler / no signup process required, no link to social networks, better sharing with third party services), but you should make a good case on why users would like to use your service rather than going to Google.

I interviewed many candidates with PhDs for one of those major software firms, and I was not long ago one of those candidates myself.

My advice is to spend some time and practice interview questions: go from the problem, to the idea of the solution, AND WRITE DOWN THE IMPLEMENTATION in a piece of paper (or screen).

The last step is the key. Many people I interviewed knew how to solve the problem, but then had no clue on where to start when writing down a concrete solution to it. This is not a matter of being not technical, it's really a matter of practice.

Also, in those large firms, nobody cares that you already know a bunch of languages. It's better being able to demonstrate that you know one of those well. Part of the assumption is that if you know one well, you can learn the others if needed.

My advice is to chat with previous and current students to get a feeling on how it is to work with that professor. In this way, you get a view of their experience.

Is the PhD in the US, Europe, or somewhere else?

As for government funded: you also get a monthly stipend, right? How long is the PhD in your country?

Also, it's a paid position, right? Never go if it is not payed, it's not worth it.

That's a very good goal. You clearly need a better managers and better companies to work for. The research experience of the PhD might help, but it's not the solution to all problems.

For a more pragmatic point of view, look at where your future professor's students are. In a 250+ ranked school, the school name is not going to land you a great job by default. You'll need to be working with a good professor. Your summary of the publications of who you are working with doesn't seem stellar, but without details on where he publishes I can't say much. How many students does he have?

Do you have only that one offer or more?

Is it in a good school, and are you working with somebody who is publishing in top conferences or journals? If not, then you won't get much out of your PhD.

Why did you always wanted to have a PhD? Do you want to work in research or academia?

I have been using LaTex for many years and I like it, but what's stopping other document editing software (like Word) to reach the same level of quality of results? LaTex has been around forever, but nobody else is close to that quality.