bendyBus

"If your task can be expressed as a workflow, build a workflow". 100% true but the thing all these 'agent pattern' or 'workflow' diagrams miss is that real tasks require back-and-forth with a user, not just a Rube Goldberg machine that gets triggered in response to a _single user message_. What you need is not 'tool use' but something like 'process use'. This is what we did at Rasa, giving you a declarative way to define multi-step processes. An LLM lets you have a fluent conversation, but the execution of the task is pre-defined and deterministic: https://rasa.com/docs/learn/concepts/calm/ The fact that every framework starts with a `while` loop around an LLM and then duct-tapes on some "guardrails" betrays a lack of imagination.

This is nice. Pulling data from an API and putting it in a SQL database should be a simple everyday task but the tools for this are 99% massive overkill. Great to see a simple library for a simple job.

The Rasa Core framework has been in development for 1.5 years, so we're incredibly excited to launch it & make it accessible to everyone.

We wanted to make Machine-Learning based dialogue usable for as many developers as possible. There's so much great research in conversational AI but very little that developers can actually use to build bots and voice apps.

One of the big challenges was making Rasa Core useful on day one, when you have no training data. We came up with the interactive learning approach, where you teach the system by talking to it and providing feedback. Rasa lets you scale from nothing up to a robust dialogue system trained on a large number of real conversations.

We're so happy with the community that's developed around Rasa NLU, and so we're really, really excited to see what people build with Core.

we could integrate with other backends, including NLTK & coreNLP. The stanford stuff is under GPL though, prefer to promote startup-friendly licenses.

yes! Docker Cloud isn't quite working yet but the Dockerfile should work :)

currently there are no built-in entities like dates, times, locations etc. But really keen to set up a way for users to share models, and that would definitely include these things as well

yeah you're quite right, intents are built with a bag of words model & doesn't take order into account. Entity extraction does though. If you find a case where word order is really important for getting intents right I'd love to know about it! We could find a way to make that work.

I'm gonna clean up the code a little bit and then make it public. It's written in scala using Play! framework and Akka.

and aloo how was the search quality? We noticed delays for up to 0.5s today because of heavy traffic. When things get quieter that should improve dramatically. Cool input on the onboarding. We found it took quite a bit of work to make a really slick onboarding flow, but there's clearly still room for improvement.

will check that out! thanks

Everything non-linear is linear to a first-order Taylor expansion. Hence why you can evolve small perturbations, the same thing is used for explicit numerical integration of non-linear equations.

I am indeed concentrating on the types of companies YC would invest in. I don't know anything about lifestyle businesses but what you say sounds reasonable.

I would assert that a recent graduate who wants to be an entrepreneur is mostly doing themselves a disservice by studying those things. It focuses your thoughts on how to run a business rather than how to solve a problem or turn an industry on its head.

Sure, many of those things could come in handy in the future - but there are far more difficult things which you also have to learn "as you go".

Point taken. Education is of course a valuable thing - but a year's worth of instruction on entrepreneurship after you've already got a degree?

I would suspect that this is non-obvious to people who view YC as a structured, semi-guaranteed route to success, the way many people view a degree from a prestigious university. I know many people who are 'interested' in startups and crave some kind of validation that they are a real entrepreneur before sinking their teeth in. My university even offers a 1 year diploma in entrepreneurship. Madness! As though formal training will help you make the hard decisions you face as a founder.

(1) Surely culture and technology stack are reasonably independent of the dimensions considered here (2) Wouldn't you agree that this asks an important question which is currently absent from the hiring conversation? (3) Even if the spider plot is an imperfect representation, isn't it a helpful starting point for a conversation between a potential employer and employee? (4) I'm not sure what your concern is about toy models and management. All models are toy models, the only thing which matters is whether they lead you to ask the right questions.

Good point, but I could imagine certain shapes becoming very recognisable after a while. Ensuring the ordering is such that a 5-1-5-1-5...pattern is unlikely, since these spiky shapes will all look the same. And yes the angles don't carry any extra information, but they are a helpful guide to the eye. I can see where you're coming from, but thinking about these plots a bit more carefully I think you are too dismissive.

I've had to spend money at various points getting documents drafted which, like this, are pretty much cut & paste anyway. I would just like to see MORE of this. Recently came across lawbite https://www.lawbite.co.uk/ which looks useful at least for UK based companies. Haven't used the service so cannot endorse. Basic employment contracts, vesting schedules, t&cs, keep them coming!

What can I say? our bots just really hit it off.

Would you be willing to elaborate re my reply to colinramsay?

"learn once, write everywhere" - I like the sound of that! But I guess in the spirit of React this is still only the "V" in MVC right? I don't suppose they are planning to broaden the scope to anything other than rendering views. So if you want to use this to build native apps, I guess you are going to have to integrate another cross-platform solution for storage/sensor APIs?

I'm not sure I understand how it's different from YAML.

but really - and I'm sure this is your `secret sauce' so you don't want to give away too much - how do you get hypers which just work out the box? was this some kind of meta-regression on the hypers? Or did you do a bayesian optimisation?

Yes, I'm part of the GAP research group (with Albert Bartok Partay and Gabor Csanyi). Representing environments is definitely still an ongoing research project. Other things I've played with in the past include identifying crystal structures at finite temperature (i.e. a classification rather than regression task), or differentiating between amorphous phases (since e.g. water has two amorphous solid phases with a 1st order transition in between, but there is no way in hell you would be able to tell one from the other visually.

We're currently working on a really ambitious new way to represent environments, but it's really preliminary at the moment.

Regarding your ion issue, what about the angular components? The radial functions really only tell you so much...

But more fundamentally, what do you mean by ion energy levels? I'm presuming you mean a metallic nucleus+core electrons, in a condensed phase at finite temperature. But of course that `atomic energy' -insofar as it exists- is a continuous function of position and not quantised, so I'm unsure what you mean by energy levels in this context.

Have been following these guys for a while and had a lot of fun playing with their API when they were still in the alpha stage. Really impressive stuff being able to do NLP in a black-box fashion. I'm sure it took a lot of time getting the default machine learning params to work well in so many cases. Admittedly haven't tried any competitor's products yet so would be keen to hear from those who have how it compares.

Sanity disclaimer: I'm sure right/left brained isn't the best or most current model of personality/cognition, but it's widely known and helps to frame this question, which I find interesting.

Is there a personality type which makes for better programmers?

Character of the archetypical left-brained person: fastidious strong logical/reasoning skills thinks in terms of structures

whereas the right-brained person finds creative/un-obvious solutions to problems good at thinking laterally thinks in analogies, better at spotting similarities than differences

Now there are many different ‘modes of thought’ a programmer encounters: coming up with the organising principles of a framework, chasing down a Heisenbug, finding a (slightly dirty) solution which saves having to re-write masses of code; these require very different cognitive skills.

Do different parts of a companies’ dev community tend to be populated by one personality type? Should even small teams contain a mix?

Is it possible to be a very right-brained, very productive programmer? And if so, is it at all clear to non-programmers that a career in software development is possible if you’re the “creative type”?

I'm not sure if I would agree with categorising the field as a race. I absolutely agree that the number of people using Machine Learning within the atomistic simulation community is skyrocketing. But everyone is just exploring the range of possibilities and trying to see what the essential elements are for it to be successful. I think having more people working on it is a great thing!

Regarding LAMMPS, actually the GAP code is now also easily usable there with this plugin : https://github.com/libAtoms/QUIPforLAMMPS

The bispectrum is indeed a very powerful tool, but is not the ideal feature vector for representing the atomic environment. You should have a read of Bartok's more recent paper on this: http://journals.aps.org/prb/abstract/10.1103/PhysRevB.87.184... . One of the issues is that the bispectrum starts with an approximation of the neighbourhood atomic density as a sum of delta functions. Trying to represent such sharp features in a basis set expansion is actually very slowly converging. So the idea behind SOAP is to build a covariance kernel by directly comparing a smooth measure of the similarity of environments, which is also invariant to all physically relevant symmetry operations.

I would also like to add that in addition to GAP and SNAP, there are people like Jörg Behler doing this with Neural Networks and Francesco Paesani/Greg Medders with a different regression schemes. But in addition to making potential energy surfaces there are people like Paul Popelier `learning' atomic charges for building force fields and people in Vijay Pande's group doing machine learning on MD trajectories, which is something that excites me a great deal and I would love to understand in more detail.

It's a very exciting time to be in this field!

There is a lot of hype surrounding Deep Neural Networks which seem to be able to solve extremely challenging machine learning problems with almost no human tuning of parameters.

This is an informal talk at the big-O meetup in London discussing the application of machine learning to quantum mechanical simulations of atoms and molecules. This is a particularly demanding application of machine learning techniques. The requirements for regression accuracy are very high, and in addition a number of physical laws need to be obeyed by the learning algorithm. The point of the talk is to invite discussion about the relative merits of domain expertise versus general-purpose algorithms for high-performance machine learning.

would like to share this which I found on twitter recently:

The Law of printf debugging: debugging messages inserted to track down unwanted behavior asymptotically approach "o_O"

-- @a_cowley

So thanks alixander for a considerable step in the right direction.

8/ A new generation of apps redefining work habits is well overdue. The obvious driver I see is that people's expectations for software usability/design are set by consumer apps. Also millenials are entering the workforce, and they have never had to take a course to learn how to use a piece of software. Any tool which requires training is arcane in their eyes, and probably rightly so. Add zero downtime, access-and-share from any device, auto sync & backup, beautiful UI to the list of requirements. Some of these are being addressed by office 365 and the like, but unbundled web-based productivity software looks ripe to dethrone mircosoft as THE enterprise productivity software vendor.