5 | ProbR – Part 2

We’re at the ProbR Launch [LIVE BLOG]

Much-hyped startup ProbR is holding an event this morning in San Francisco to announce the beta version of it’s long-awaited “AI answer engine”.

We’ll be live-blogging the event and will update this post as we receive information about ProbR, insights from the company’s founder Michael Probst, and any images of the new site.

Stay tuned for more.


[All times are in Pacific Standard Time.]


10:30 a.m. They’re starting on time! By way of a teaser they show us a screenshot of ProbR.

They’ve typed the query terms “Probst” and “mean” and are showing us the results page, which has three entries:

1) “The name Probst comes from the Middle High German probest, meaning ‘supervisor’ or ‘provost’ (from the Latin propositus)”

2) “The average income of people with the name Probst is $86,243” (and conversions in Euros and Yuan)

3) “The term Probst appeared on average .000149 times daily in real-time search results last month”

{I feel like I’m at the movies watching preshow trivia. The answer’s Matt LeBlanc. Isn’t it always Matt LeBlanc??? Can’t they make the questions harder?!}

10:35 a.m. Marsha Maier, Chief Insurgency Officer. Five years ago ProbR was just a gleam in it’s founders’ eye. Probst thought the only real Google killer would be something that could intuit user’s intentions and “give them a little window into the future.” Knowing something even five minutes before its going to happen yields huge competitive advantage. ProbR will deliver on this promise in ways no one’s been able to do before.

10:40 a.m. Ahjit Shukla, Director of Research. Shukla explains that from the start, the purpose of the program was to give predictions about future events and phenomena.

A multinational team of researchers was put together to start “feeding the beast” with curated data which they used to design the first version of the AI interface. Early answers resembled Google search in randomness and computation time could run up to 30 seconds.

Shukla tells an old algorithm joke: “Prove: Given any algorithm, you can always make it run more slowly. Is the converse true?”

10:45 a.m. Shukla explains that they started from a base of “known knowledge” and created an average search engine capable of returning answers to simple arithmetical, geographical, or temporal queries. ProbR draws from nearly 20 trillion pieces of discrete information.

He says ProbR’s technical team combed through text, image, and video files—“mostly text, though”—to create an associative data array. “We had to find a way to take the aleatory nature of knowledge and turn it into algorithms so that it would be possible to generate relevant intelligence. We started by asking, ‘what do we already know?’ and then tried to imagine what might we want to know about the future.”

11:05 a.m. Eric Dean, VP for Product Development takes the stage. Dean talks about licensing and partnerships. ProbR already has a waiting list of potential institutional partners including Altria, a Department of Defense research unit [anyone else think this is something besides DARPA?}, HSBC, Merck, and Virgin Airlines.

He says they plan to start with a white-label product and eventually roll out a free consumer platform with advertising. If you don’t want stuff blinking at you there’ll be a subscription model too.

11:05 a.m. Dean says this is the first public demo they’ve ever done, and apologizes for taking advantage of the moment to answer his own burning personal question. The screen has a live view of ProbR. Looks a lot like early Google. [We’ll get screenshots up as soon as they’re available.] Dean toggles on voice recognition and asks it who’s going to win the NCAA tournament.

According to the results the computation took 0.8 seconds. Men’s Division: North Carolina over Michigan State by 11 points, ± 2 points. Exactly 2.4 seconds later we’re looking at results for the next two seasons, although those don’t come with a point spread. Dean pulls out his tab and says he needs to make some adjustments to his fantasy team.

11:10 a.m. Finally the man himself. Probst takes the stage and gives some historical context. He saws when he wrote the program for his stats package R in 1990 his Mac G4 was running at about 3.2 GFLOPS. [Non-Geek translation: crazy slow.] Even he’s been “astounded by Moore’s leaps” in the meantime and it was this possibility that convinced him he could pull off ProbR.

11:15 a.m. Probst makes fun of Dean for choosing something so easy but says he’ll break down the process behind the query anyway. Analysis of the background information was a straightforward task since historical tournament results and player statistics are all in the database. The trickier part according to Probst is linguistic and sentiment analysis of current data coming in.

In this example he says ProbR’s algorithms aggregated all of the media stories, sports show transcripts, streaming chatter from multiple services, betting pools, high school player’s stats, recruitment patterns, draft history, and “probably several thousand other sources of data exhaust” and came up with what the results we’re looking at.

11:25 a.m. Shukla joins Probst onstage and says he wants to ask ProbR about the ICC World Cup outcome.

Shukla adds that the computations can take into account 60 languages but that they’re aiming for twice that by the end of the year.

11:30 a.m. Dean and Maier are back. Dean says his friends from business school have been following the ProbR launch via livestream and already kicked him out of their fantasy league.

Maier says the consumer product expects to roll out by the end of the year and will be browser-based. Some searches will be restricted if the intelligence involved is deemed “too sensitive,” based on the advice of trusted consultants”. Probst interrupts to say that he hasn’t forgotten his roots though, referring to R’s open-source software, and that as much of the information will be available to the public as is possible. He promises to “err on the side of risk”.

11:35 p.m. Probst thanks everyone for coming. Instead of Q&A he says a special version of the site is live at beta.ProbR.com and anyone can ask it questions about the company.

It’s over! We’ll continue to cover developments about ProbR and in the meantime leave us comments below. Thanks for joining us.



Image Credits (all via Creative Commons)
calculator: Dave Dugdale
pie: Elly Jonez
sushi: Zoe Shuttleworth
pasta: Steve Johnson
Rob: Chris Betcher
Nick: Colin Davis
Amrit: Mike Goren
Cameron: Mohammad Fahmi Mohd Shah
Mike: Allanah King
Xeni: Nic Rad
Curtis: Brian Rosner
Samir: Adam Schuster
Liz: Deborah Morrison
Che: Ben Sutherland


Gimme Feedback! 1) The ads are intended to be annoying (as is the bad Photoshop). Did it work? 2) If you have ideas for ridiculous startups, I wanna hear 'em!

Comments are disabled for this post