jump to navigation

Refresh the Old and Tired (Notes from July 8 to 14, 2019) July 30, 2019

Posted by Anthony in Automation, Digital, experience, finance, Founders, Leadership, marketing, questions, social, Uncategorized, WomenInWork.
Tags: , , , , , , , , , , , , , , , ,
trackback

For the abundant discussion on big tech, rise of tech and the valley’s obsession with all of it, there are quite a few industries that have had much longer staying power. They’ve proved their worth, decades and decades in. There are still railways. There are still cars. Manufacturing persists. CPG and everything that that entails last. Walmart, as much as people love (or don’t) Amazon, it’s still a lion’s share of commerce. Tech has improved and allowed them to have this staying power. Additionally, enabling improved efficiencies can allow new players in the industries to fundamentally change how they’re viewed.

Industries include tv – nonpartisan and bipartisan news with Carrie Sheffield. a16z gets into online from offline forms of services, restaurants to tech-enabled deliveries, as well as the rise of CAA and the agency fights. Then we have traffic and building with a consultant in that space. The next industry was making the legal space a little more transparent – provide a marketplace where information becomes symmetrical. I believe these are ways that simple pain points that can be improved through a technological lens give access to a value that wasn’t there before.

Hope you enjoy the shorter posting and the notes as more detailed. Check each of the wonderful people out!

  • Carrie Sheffield (@carriesheffield), co-founder of Bold TV (Wharton XM)
    slack-for-ios-upload-1

    • Discussing bipartisan vs nonpartisan
    • Growing up in very conservative areas and then going to the coast – seeing both sides, especially media
      • How it was to be in media
    • Fake news as non-fact-checked as well as actually fake – ~70%+ considering bias
    • Intellectual diversity along with everything else – thinking differently vs looking diverse
      • Used example of Google AI conference canceling on a colleague who was a conservative, black woman
  • Chia Chin Lee, CEO of BigBox VR (Wharton XM)
    ravlfjtl_400x400
  • Initially trying VR and finding it sickening – didn’t work (Oculus)
    • Tried HTC Vive and fell in love – had a room set up and felt enthralled
    • Hardware and platform may get cheaper with tech
      • Opportunity lies in the software side – connecting to others and industries

 

  • Entrepreneurs, Then and Now (a16z 6/29/19)
    • With Marc, Ben, Stewart Butterfield (@stewart)
    • 10 year anniversary for a16z in late June – how has the environment changed?
    • Class of 2009 entrepreneurs were some of the most special: Todd McKinnon, Martin, Brian Czesky
      • To get to that point, needed to earn your stripes
    • O2O – online to offline (AirBNB, Uber, DoorDash, Postmates, etc….)
      • Founders that may be more operationally-focused since those require that
        • Maybe more similar to semiconductor founders from the 1970s, start of 80s
    • Dual discipline people as they got more involved in healthcare or bio-related
      • 10 years ago, Bio PhD wouldn’t know much on computers but now, dual PhD’s
    • Economics + CS – discussion of field of economics with empirical / quantitative economics compared to physics or formulas
      • New inventions by economists with machine learning and data
    • New ideas – thought venture firms had lost way, founders/operators that built businesses who would help out on boards
      • GPs started to get more abstract ideas, professionalized
      • Institution and ecosystem, network and fundamental staffing model – pay at a16z is different than other VC’s
    • If priority was to find best founders at the best opportunities, shouldn’t matter which stage they’re at – miss things, maybe
      • Skype deal early, multiple entry points – working with entrepreneur and being stage-agnostic
      • Tech bubble bursting – “can’t possibly start fund” – 2009 was Khosla and them
        • Mentioned ‘crusty’ or ‘grouchy’ VC’s
    • Much of the tech was at an inflection point – Salesforce as only SaaS, iPhone not quite there yet, Uber, Airbnb
      • Maybe the main response should be “No, this thing is stupid” as more accurate
      • Never thought it was a bubble – prices of companies are always incorrect (future performance, which nobody knows)
      • East coast vs West coast – not obvious, find what each argue about
    • How high is up? Online pet delivery, all actually happening
      • What are the exploratory bets? Are markets ready? Are people ready? Regulators?
        • Sometimes it’s the pioneer, sometimes it’s the last – time and effort for founders, personality, other
    • No individual company gets 25 years to prove something – maybe 5 years for a hypothesis
      • Morale issue losing faith or architecture issue – prior architecture (ex: mobile dev in 2002, system on archaic and aging-in-place)
      • VC’s will do the same thing – kid doesn’t know about failed experiments – VC freeze themselves out (ones who don’t know will often invest)
        • Can you learn lessons from failure – maybe you should learn nothing – “That doesn’t work.”
        • Edison as trying 3000 combinations before the filament, Wright brothers trying many
    • Copying the model from CAA – Michael Lovitz and describing the whole thing – not a collection of individuals
      • Operating platform, system and infrastructure with professionals across the network
      • Compounding advantage year over year – but why can’t they copy? They were paying themselves all the money
        • Nobody wanted to take pay cuts – 80% to hire everyone at such a scale
    • Top end venture investment – need something working (product-market fit, product)
      • Do they know what they’re doing? Can they do their job scaling?
      • Second-time or later founders – can do what they want and figure stuff out?
        • Problem may be with the good idea – investments on that idea or otherwise (fragmented idea with nothing)
      • Idea maze to find out what the ideas are – haven’t gone through that
    • VCs can’t invest more than 20% of funds that aren’t primary equity investments – crypto, for instance (vs RIA)
    • Deadwood as creation of city or state – horrifying obstacles
      • Why History is Always Wrong? (Taleb’s narrative fallacy, for instance – often more complex)
        • Don’t even know body, climate still (too complex) – can converge on science to Newton’s laws, others
      • Can’t Hurt Me by David Goggins
  • Scott Kuznicki, Pres and Managing Engineer at Modern Traffic Consultants (Wharton XM)
    logo-text

    • Traffic control tech – California high speed rail vs autobahn style
      • Autonomous lanes?
    • Designated autonomous – level V vs others, depends on density and adoption
    • Thinks parking structures with flat tops could be converted or pay for cost
      • Multipurpose, solar, green or plants etc…
  • Risk, Incentive and Opportunity in Starting Co (FF 027, 20min VC)
    logo-062fd8699c93a47b2e8278975e71b84870194fd30c288607f1e06c92a4e831a0

    • Daniel van Binsbergen, CEO and co-founder of Lexoo
      • Online marketplace connecting businesses and lawyers
    • Founded it in 2014, got an investment for $1.7mil
    • Friends always asking for referrals – kept a short list of them
      • Seemed great, “quoted $X – is that good?” – perception of complexities
      • Could put make a marketplace together for transparency
    • Kept 100% of his income boosts – got used to his training salary so it wasn’t as big a risk
      • No kids meant it may have been easier – really disappointed if you didn’t give it a go – decision already made
    • Legal space’s lack of progression in tech – incentives in wrong place
      • Hourly model still for law – if you spend less time on work, you would make less money
      • Risk-spotting for lawyers
      • Senior partners have heaviest voice – not exactly lining up for retirement in the near term vs long term
    • Highest goal may not be senior partner – fixed fee, sharing risk, more open to innovating with own practice
    • Lexoo initially – didn’t have tech skills for it, had a vision in his head but didn’t know best way
      • Didn’t build full-scale solution, did a forum for $15 website, form to fill in
      • Arrived in his email – he would then contact lawyers and fill in Word template – get their responses and quotes
      • Attached the lawyers’ quote and response to a doc and pdf and send back to clients
      • Automated only when he couldn’t handle the workload – hit limit on evenings and quit
        • Lawyers paid 10% commission on the quotes
    • Focus on business ideas – tech isn’t the big solution – market innovation (access to litigators)
    • Investors at Forward Investors – introduced through a friend who knew them through squash partner
      • Difference between FOMO on being convinced vs other investors who have a sense of opportunity
    • Fav book: The Mob Test – how to ask questions to get useful feedback, asking questions to customers in the wrong way
      • Would you use the product if it does X, Y, Z – most definitely? Instead of asking what the customer problems are.
    • A lot of work in Trello, for goals, and Sunrise app – Microsoft’s indispensable for calendar meetings
  • Facebook Bargaining Bots Invented a Language (Data Skeptic 6/21/19)
    • Auction theory and econometrics – equilibrium strategy
    • Neither agent is incentivized to change strategy if the other stays the same
    • Plateau of events in real life – baby, marriage, life changes, job, lease ends in time
    • Discount is a single floating-point decimal, ex 0.99 ^ t
      • Everything known – can calculate based on common knowledge and discounts
    • Gaussian distribution, mean 100k, 10k – ignore tail in negative and renormalize
      • Rubenstein one-sided incomplete
    • Game: don’t know private value now, but can have probability distribution
      • Update with Bayesian with behavior
      • Classic ML: corpus of examples of negotiation, mark up conveniently, objective function to maximize reward (post-agree)
      • Opportunity for RL – patterns for language utterances, insult or compliment or neither – recognizing strategy
        • Character level or nothing to ask it
        • Conversations for language you don’t understand and the reward – can you do this optimally?
    • RL + Roll-out with 8.3 to agent and 4.3 to other algorithms (94.4% agreement)
      • Roll-out was 7.3 and then RL – 7.1 and last place was 5.4 for likelihood model
    • Training data was in English, negotiating over 3 items – shortcut its job, RL wants the short path to reward
      • His example – loses points if you went to pits but to reward – chance at falling
      • Wasn’t worth it to move, so he had to do a penalty for not moving
      • Penalty for Facebook example was agents continued to communicate in English
      • Put a time constraint, maybe
  • Transfer Learning with Sebastian Ruder (@seb_ruder), D/S at DeepMind (Data Skeptic 7/8/19)
    deepmind-1

    • Generally, TL is leveraging knowledge from different tasks or domains to do better on another task
    • Not a lot of training data, may want to pretrain – models to train on imagenet, for instance
      • Language modeling to train on large corpora and use that on a bunch of other tasks
      • Source vs target data: task stays the same but can adapt between source and target, say sentiment of reviews
    • Classic benchmarking, may have ImageNet moments over last year – features of pretrained models applied on more powerful NLP
    • Google XLNet’s most current, BERT and ELMo as others – pace of improvement has been great
    • Difficulty of target tasks – can be good for 100 samples in target source on binary tasks, maybe, 50 even?
      • 200 examples per label, question-answering or reasoning, examples must be increased
      • If we can express target task as a conditional language modeling, can do fewer or even inference
    • Pretraining is costly due to large clusters on your own, but now can be public pretraining where you can finetune quickly
    • Area of common sense reasoning – infer what a question means or expressed depends on what may not be said
      • Grass is green, entity facts (son of a son), inquiries for language model – incorporate to modeling
Advertisements

Comments»

No comments yet — be the first.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: