## Best Live Chat Widget for Startups: Jan 2016

January 20, 2016 § Leave a comment

Live chat support software is a great way to ensure visitors to your SaaS site are successful – or at least tell you what went wrong. An in-page widget with a friendly human being on the other side is much more approachable than a link to anonymous forum!

Unfortunately, few startups are able to guarantee 24×7 coverage, which means you need a easy-to-use tool with a good fallback user experience for when nobody is available to chat. Here is what appear to be the best options as of January 2016.

## How to Build a Team

November 20, 2015 § Leave a comment

**Vision**: What does success look like?-
**Humanity**: What do you need to succeed? -
**Process**: How do we ensure everyone gets what they need?

## Wide Open (or, Are You In?)

November 10, 2014 § Leave a comment

He’s my hero. THIS is how I dream of running my own projects / company.

Earlier this year I confronted the painful realization that my baby framework grew into a mature ecosystem – one I no longer had the capacity to maintain on my own. It started with dragging open issues for more than a few days, to a growing pile of sticky notes on my monitor of ideas I’d like to try, to (and most problematic) no longer remembering how big chunks of the code work.

The problem is, how to successfully move from a one-man-show to a community driven project, without giving up on the stability, consistency, and philosophy of the framework.

**Consensus-Dictator-Fork**

I believe the only practical model for running a successful open source project is the Consensus-Dictator-Fork (CDF) model. It’s a fancy name for how most open source projects work. Decisions are made by consensus whenever possible. This usually covers 95% of the decisions by the simple mechanism of proposing a…

View original post 932 more words

## This is Probably a Good Time to Say That I Don’t Believe Robots Will Eat All the Jobs …

June 17, 2014 § Leave a comment

The mechanics of jobs will be automated, which is why the jobs of the future will rely on us being more human to each other.

** **THE ROBOT TWEETSTORMS by @PMARCA

One of the most interesting topics in modern times is the “robots eat all the jobs” thesis. It boils down to this: Computers can increasingly substitute for human labor, thus displacing jobs and creating unemployment. Your job, and every job, goes to a machine.

This sort of thinking is textbook Luddism, relying on a “lump-of-labor” fallacy – the idea that there is a fixed amount of work to be done. The counterargument to a finite supply of work comes from economist Milton Friedman — Human wants and needs are infinite, which means there is always more to do. I would argue that 200 years of recent history confirms Friedman’s point of view.

If the Luddites had it wrong in the early 19^{th} century, the only way their line of reasoning works today is if you believe this time is…

View original post 1,771 more words

## How to Have and Resolve An Argument

June 15, 2014 § Leave a comment

**Premise**

We hold certain positions because of what we:

A. Experience -> B. Encode -> C. Evaluate -> D. Emphasize -> E. Express

## The Multi-Minion Machine

August 9, 2013 § Leave a comment

### A Function of Scale

### Draft 1, Ernest Prabhakar, 2013-08-08

### The Sequel to “The Minion Machine“

## The Premise

Real systems aren’t linear, but have scales where the cost is fixed below, but astronomical above.

## The Goal

Extend/Restrict the Minion Machine to capture what it means to operate at “optimal scale”.

## The Concept

Define a **Multi-Minion Machine** as a Minion Machine with the following changes:

- There is one minion for each bin (and thus each object)
*(M = N)* - Minions never move; they just shoot objects to other minions.
- The
*N*objects are arranged in a ring of radius R, so “1” is next to “N”. - The objects travel on independent tracks of size r << R, so they don’t collide, but take effectively the same distance to a given bin.

## The Model

Assume the minions are smart enough to figure out the optimal route from one bin to another. Instead of specifying a distance, we can thus just specify a destination (and not have to worry about ‘overflow’ or ‘underflow’).

Our primitive commands only need specify the initial (*b_i*) and final (*b_f*) bins, giving a size of:

*S1 = 2 log(N) := 2 k*

All other quantities are the same, except that the average distance *d* will be less (half?) due to the ring topology.

## Metrics

Let us use **bold** characters to represent an action tuple *(E, t)* whose norm is *E times t*. For example, operation **L** has the action *A_L = (E_L, t_L)*. The action of our system can be decomposed into **C** for the communicator and **M** for movement.

If solving the puzzle requires *n* commands of size *S1* and average distance *d*, we can write our action as:

**A0** = n S1 **C** + n **M(d)**

[: parallel operations could complete in a time proportion toErratamax(d), independent ofn. There is a complex dependency on the relative values ofC_tandM_twhich I overlooked].

Now we can ask: would higher order commands reduce the action?

To start, let us introduce a program with per-command cost **T** that interprets a command as a transposition instead of a move. For example, if *N = 8*, the command 0x1f is split into 0x1f and 0xf1 and executed in parallel.

For a set of disjoint transpositions that would normally take *n* moves to solve, the action is now:

**A1** = n/2 S1 **C** + n/2 **M(d)** + n/2 **T**

For this case, it is a net win when (substituting *k = log(N) = S1 / 2*):

**T** < 2 k **C** + **M(d)**

which is a net win for sufficiently large *k*.

However, that advantage only holds for disjoint permutations. Conjoined permutations (e.g., cycles) take the same number of steps as before, but most now pay the penalty**T**.

To solve that, we could replace **T** with a program **L** that describes loops (cycles) rather than mere transpositions. This gives us, for all (?) permutations:

**A2** = n/2 S1 **C** + n/2 **M(d)** + n/2 **L**

with a similar constraint:

**L** < 2 k **C** + **M(d)**

## Interpretation

A particular command/program specification can be interpreted as a “strategy”.

For example [as Christy suggested], imagine two players Satan and God.

- Each of them is given a Multi-Minion box for which they devise a fixed strategy behind closed doors.
- When the curtain comes up, Satan & God get to see each other’s strategies.
- Satan secretly feeds commands into his box to entangle a set of balls.
- Those balls are teleported into God’s box, where he must dis-entangle them.

Every command costs some number of “action points” (great name, Christy :-). The winner is the player who spends the fewest action points.

This leads to a number of interesting questions:

## Questions

- Are there optimal strategies for God and Satan? Is the optimal strategy the same for both players? Is there a meta-strategy for which commands Satan should use, after finding out God’s strategy?
- Does one player have an intrinsic advantage in this case? What about the case where the entanglement isn’t simple permutations, but some NP-complete problem?
- How should we calculate the per-command cost
**P**for the program used to implement the strategy? Naively,**L**ought to be bigger than**T**, but by how much? Can we break all possible strategies down into a “basis” of simpler components, allowing cost comparisons between them? - Do any of these results change in interesting ways if we add baseline costs for any of the elements?

## Conclusion

I’m not sure if we learned anything about scale, but we did develop a useful concept of strategy. It also implies that the action (which is perhaps closer to “difficulty” rather than mere “complexity”) depends on interactions between the instruction set chosen and details of the input vectors.

Then again, maybe that is *why* we have different scales: to allow optimal instruction sets for different levels of representing a problem…

## The Minion Machine

August 9, 2013 § 1 Comment

# The Action of Complexity

### Draft 2, Ernest Prabhakar, 2013-08-07

### Inspired by a proposal from Christy Warren

## The Premise

Using concepts derived from physics such as Energy and Time, we can gain insight into the nature of computational complexity.

## The Goal

Devise the simplest possible physical system that captures the aspects of computation relevant to complexity theory.

## The Concept

The goal is to move the objects from an initial ordering *I* to final ordering *F* while consuming the least amount of time and energy. Importantly, the only way to accomplish this task is by giving commands to the minions. Minions only understand commands of the form “Minion – Starting Bin – Direction – Distance”.

## The Model

We start by making a number of simplifying assumptions. These can be revisited later as needed.

- The energy required for the minions to live and move themselves is either negligible or from an external source. The only energy we care about is that required to i) move the objects and ii) communicate with the minions.
- All objects have significant mass (so it takes energy to move them) but negligible size (so we don’t need to worry about collisions).
- All objects have the same mass
*m0*and top speed*v0*. The array has negligible friction, and the distance between bins is very large compared to the distance required to accelerate to top speed. This allows us to assume that moving any object from one bin to another takes the same amount of energy (to accelerate & decelerate):

*E0 = m0 v0^2*

but a varying amount of time, proportional to the distance *x*:

*t = x / v0*

- The communicator uses something like FM modulation, which requires energy
*E_c*and time*t_c*both proportional to the dimensionless size*S*of the command, e.g. in bits:

*E_c = a S t_c = b S*

The sizes N and M are fixed, so we can specify that all primitive commands use a fixed-width bitfield of size *S0*:

*S0 = log(M)+ log(N) + 1 + log(N)*

## Metrics

Say that it takes *n* steps to obtain the desired order. The distance traversed by each step is given by *x_i*, which can be summed and divided by *n* to get the average distance *d*.

Assuming serialized movements with no latency between them gives:

*Energy = n a S0 + n E0 = n (a S0 + E0)*

*Time = n b S0 + n d / v0 = n (S0 + d/v0)*

We can multiply these to get the action:

*Action = Energy * Time = n^2 (a S0 + E0 ) (b S0 + d/v0)*

Since S0 is dimensionless, we can pull all the dimensions in a new constant *h*, which is the per bit action of the communicator:

*h = a b*

giving us new dimensionless constants:

*e = E0 / a f = d / (b v0)*

allowing us to write:

*Action = n^2 h (S0 + e)(S0 + f) = n^2 h (S0^2 + 2(e + f)S0 + e f)*

## Interpretation

The action can be interpreted as a measure of the effort required to ‘disentangle’ a system from an initial ordering *I* to final ordering *F*.

The constant *e* is the ratio between the energy required for each step of movement (*E0*) and that for each bit of control (*a*).

The constant *f* is the ratio between the average time required for each step of movement (*d/v0*) and that to send each bit of control (*b*).

Which tells us that the effort is primarily determined by:

- Movement, when
*e + f >> 1* - Control, when
*e + f << 1* - Energy, when
*e >> f* - Time, when
*f >> e*

While those are perhaps obvious, this model also provides a precise way to measure the effort (action) in intermediate cases where *e* and *f* are comparable to 1 and each other. It also gives us a mathematical formalism that can be used to minimize the action when varying some of the constants or extending the action.

## Questions

- Is the action the right way to combine
*E*and*t*? What are the alternatives, and their advantages and disadvantages? - Right now having more minions doesn’t help (or hurt). What happens if we include their energy cost, but allow them to perform actions in parallel? What if we are allowed (at some cost) to send the same command to multiple minions at once?
- What if the energy cost is dependent the distance between bins, rather than constant?
- What is the physical interpretation of
*h e f*, the “pure movement” action*d E0 / v0*?

## Conclusion

This model leads to a natural and interesting definition of action for computational systems that bears some interesting similarities to the idea of ‘complexity’. To flesh this out, however, would require a mechanism for encoding (and costing) higher-order algorithms such as “sort the array”, rather than merely “move these objects between bins”.