janhenr

Converge (YC S23) | Founding Account Executive | London | £80-100k

Converge is vertically integrating the modern data stack for B2C by bringing customer data collection, identity resolution, attribution, data forwarding, and analytics into a single platform.

We are urgently looking for our first sales hire. This role consists of owning sales cycles end-to-end, helping us build out sales enablement materials, and listening deeply to customer problems to solve

This is a very exciting time to join Converge:

- 0 -> 160+ customers in the last 18 months

- revenue 5x in the last 12 months

- cash-flow positive in the last 8 months

- more than doubled usage in the last 4 months

Full role desc here: https://jobs.gem.com/converge/am9icG9zdDqATabxB8rqC-XFCYn_kh...

Converge | Founding Product Engineer | London | Full-time

Converge is building data infrastructure for online stores. We make it easy for merchants to understand where their customers come from and help them optimize their customer acquisition.

We just graduated from the YC S23 cohort. We serve more than 100 customers (small stores until publicly listed co's) and are profitable.

We are looking for a full-stack engineer who gets excited by talking to users and shipping product with us. You will be highly autonomous as to what you choose to work on to maximize your impact on the company.

Apply here: https://www.ycombinator.com/companies/converge/jobs/vc4Nbeh-...

This is not entirely correct, many places still do accept cash and allow you to order without your phone. It just so happens that most people prefer to order and pay by phone.

This is where it helps to have a diversified cap table. In our case, we have a set of angels -- each with their own expertise.

All of them lead busy lives, but we can get many hours of feedback by picking a subset to talk to on a particular subject. E.g. we have a "former CTO of large SaaS company"-angel who we specifically call with regards to tech stack; an "ex Head of Product" we ask for feedback on product etc.

As an added advantage, it almost automatically leads to strong mentor-mentee relationships on a personal level.

Totally understand your mixed feelings. We are trying to do as many standard data preprocessing steps as we can (cleaning, normalizing, one-hot encoding etc.) and are now switching over to an AutoML engine. We basically see the outputs of Magicsheets as great baseline models, Data Scientists could definitely do better (especially with domain knowledge) but we should at least be able to give some useful predictions back for most problems :).

Thanks for the insight! We expect to keep the Google Sheets add-on free for the near future. For our business model, we are instead focusing on building fully end-to-end pipelines that use the same ML engine as for the Google Sheets add-on:

Data integration -> ML Predictions -> taking actions.

As an example, for an ecommerce store we plug into their Shopify, predict people likely to churn and send them a Mailchimp discount code.

We are running pilot projects now for these pipelines, so if you are interested do not hesitate to reach out to me on jan [at] magicsheets [dot] io

Hi there, thanks for the kind words and advice! Our next step is building an AutoML engine that indeed takes algorithm selection and hyperparameter choice out of the hands of inexperienced users.

Hi Nalta! Thank you for your advice. We noticed that indeed most spreadsheet users do not do well with ML logic - as they are used to Excel functions, which have an input -> output structure (not validation cycles, hyperparam optimization etc.). We are building an AutoML engine, exactly for this reason - to indeed take everything "ML" away for spreadsheet users. Our ultimate goal is to turn ML literally into an Excel function. Super challenging problem, but really exciting at the same time :).

Thank you so much! Please do reach out with any questions/comments etc. - super happy to help :)

You can always reach us at hello [at] magicsheets [dot] io

Hi HN,

I found it frustratingly hard to get Machine Learning baseline models on my own datasets (usually in Google Sheets) so I pulled a couple of friends in and we built a Google Sheets addon that takes care of:

- Data preprocessing (cleaning, scaling, etc.)

- Applying ML algos

- K-Fold Validation

- Returning predictions

So you can get baseline predictions in a few clicks while never leaving your Google Sheets file!

We are really excited to get some feedback from HN :).

Also, if you have any questions, shoot!

Know the founders personally, but my comment still stands. Was honestly not a fan of the whole idea up until their launch day today.

Interesting piece! Thanks for sharing. Nitpick if you are the OP: while I appreciate the distinctive style, the font really made it hard for me to read

Hi there! What you could do is give a GPA just for your major courses (which are more likely to be later in your curriculum).

Hi Bert and Otto,

As a fellow belgian, I have been following you and segments closely. Congrats on being the first belgian YC company :).

From the beginning onwards I was wondering why you chose to put such an emphasis on segmentation labelling. Do you see this usecase as the Computer Vision application with the biggest (future) market or maybe the least saturated offering at the moment?

Check out ohmyzsh for a great ~/.zshrc

I genuinely wonder what this means for the future of DeepMind. Is it supposed to be profitable or is it just R&D spending for Google? Where are the paths to profitability for them? Are they counting on more expensive GPT-4,5,6 APIs?

I guess I'm just asking if COCO or ImageNet-trained networks are actually noticeably superior for most real-world tasks, or if it's just a metric that's used because the performance differences only show up in the long tail of the distribution.

Given that for any real-world vision task you start from a pretrained model om those datasets they will in fact be noticably superior on the real world task after finetuning. Just because the quality of the features extracted through the backbone is better.

I do really like the idea of an MNIST alternative to very quickly verify ideas. However I have a few nitpicks:

1. 10 classes is way too small to make meaningful estimates as to how well a model will do on a "proper" image datatset such as COCO or ImageNet. Even a vastly more complicated dataset like CIFAR-10 does not hold up.

2. I feel like CIFAR-100 is widely used as the dataset that you envision MNIST-1D to be. Personally I found that some training methods will work very well on CIFAR-100 but not that well on ImageNet so TinyImageNet is now my go-to "verify new ideas dataset"

LPT: firefox on mobile allows you to install extensions such as ublock

The League of Legends x Apple announcement is bigger than people might think. I 100 % believe that this is the start of a massive shift from pc/console to mobile which we have seen before in China. In fact, it is clear from this announcement that iPhone IS considered a console now for Riot. Huge stuff.

I think we can both agree that there is a lot of value in knowing the "actual value" (as opposed to speculative value) of things in our society. Knowing the actual value of things is impossible for multiple reasons it (it changes very fast and is just intrinsically very hard). However, given a big incentive (making money) we can let the market estimate the value of goods for our society. This speculative valuation is far from perfect, but it is the closest we can get to the actual valiation of goods/institutions.

The fact that stocks can be bought and sold quickly has a use, namely that the price should signal how the market values this asset.

As most of Bezos' fortune is in Amazon stock, this money is not exactly sleeping.

No, in production rectangle images are reshaped and resized to square ones.

Last amount disclosed was in the area of $25B

Something is definitely moving in the market. A lot of these IPOs (Unity, Palantir, AirBnB, etc.) were highly anticipated months back already.

I enormously appreciate the resources OpenAI provides to start out in DRL such as this one. However, OpenAI has (purposely?) left out the brittleness of their algorithms to parameter choice and code-level optimizations [1] in the past. As a researcher myself, I would be more than surprised to hear that OpenAI did not explore this behaviour themselves. Instead, my guess would be that these "inconveniences" would do harm to the Marketing of OpenAI and its algos. Such deeds are far more harmful to proper understanding of DRL and applications than a nice UI is beneficial imo.

[1]https://gradientscience.org/policy_gradients_pt1/

Just fyi: On Firefox Mobile, the menu text (white) disappears into the background (white).

There might be some other reasons for shipping different models for the iPad vs the iPhone. F.e., if the iPad is more often used inside rather than outside, you could use a fine-tuned version of your big CNN to this smaller set of classes.

Well, there is a whole seperate line of research concerning the topic of these input perturbations ranging from PGD to just Gaussian noise. This model does not claim to defend against any of those.