News for hackers. Stuff that matters.

geirfreysson

I'm a bit skeptical about sending all my code to a cloud-based LLM, so I built a command-line tool that does code reviews using local LLMs via Ollama. It can also hook into OpenAI if you prefer (that gives better results tbh).

I've also had fun piping the results into other command-line tools, such as Simon Willison's llm

Re:Farewell from Rusty Russell on Wednesday February 14, 2018 @07:34AM (#16375293)
"Actually, bitcoin is a nice reward too; it's like crystalized machine sweat!" Crystalized machine sweat - very good.
Re:Poll: What OS do you currently use? on Tuesday August 13, 2013 @07:26PM (#6208993)
My other computer is a data center.
Re:Growing user signups via data and analytical thinking on Wednesday February 27, 2013 @04:25AM (#5291458)
A very interesting eBook on growing user numbers. Warning: First you'll need a product and a trickle of signups to actually optimize.
Fair point - but the whole idea is to create a light-weight CRM system without the tabs (and indeed without deals, pipeline etc). We keep it lightweight enough to be wizard-friendly which means that it's usable without investing too much time in setup. And we still use a traditional CRM for rev. forecasting etc.
The log-in screen is sparse - that's true. We're getting a (possibly slightly biased) 30% conversion rate from it, so it does push people to give it a spin. Same idea as the Twitter landing page. What you get as soon as you log in is a wizard which takes you through the first steps to maximise the probability of engagement.

37Signals wrote a very interesting post on results from their A/B tests on the landing page for Highrise (result, the shorter the better) http://37signals.com/svn/posts/2991-behind-the-scenes-ab-tes...

That's a fair point. Customer segmentation can be overdone.

There was an interesting article on HBR about research Clayton innovator's-dilemma Christensen was doing into customer segmentation vs. a job-to-be-done approach (http://hbswk.hbs.edu/item/6496.html).

Once someone clicks on a link that calls out to them as a customer, you need to tell them pretty quickly what you can do and how you can do, i.e. what job you can help them do.

On brandregard.com we simply moved the discussion about features down one level, behind a customer segment link. We also discuss features in length below the fold on the landing page.

Starting out with x and ending up with 3x is is a 200% improvement. You think "threefold increase" should mean x+3x=4x?
I agree. We're pretty sure who are customers are, but there is a chance the unknowns could slip through. We might do another A/B test simply adding "brands" as one option since that's a customer persona, but very generic.

I like how Foursquare appeal to their customer personas, "merchants", "brands", "developers", without making it too obvious but still very findable.

Exactly! We're monitoring which customer persona gets the most clicks (with good old Google Analytics). Down the line we might even try 8-10 different customer types with A/B testing and see which ones attract the most clicks.
For anyone interested we used A/B tests, powered by the seven_minute_abs Rails plug-in, to measure the results.

I'd be interested to hear about other people's experiments both with A/B tests and customer segmentation.

Re:Fake Steve Jobs on Chrome OS on Thursday July 09, 2009 @07:18AM (#695617)
Fake Steve: "First of all, nobody seems to appreciate how goddamn hard it is to make an operating system."

From the Google blog: "[GOS is Chrome running] within a new windowing system on top of a Linux kernel"

So they're not building an OS from scratch. Did Fake Steve even read the post?

Re:NY Times infographic on MJ's chart performance on Friday June 26, 2009 @10:50AM (#675491)
I know links about Michael Jackson might not belong here, but nice representation of data is always interesting ...
Also, check out the list of the most watched projects. Ruby and Rails seem to dominate there: http://github.com/popular/watched

Older »