jd007

Saris AI | Full-Stack + ML Engineers | Full-Time | Onsite / Remote | Montreal, Canada

Hi HN, we are an YC alum founding team rethinking back-office workflows for banks and credit unions. We're tackling a $100 billion problem with the kind of automation problems that require long-context reasoning, tool orchestration across many critical systems, and strict compliance loops: the ones without known answers.

We've shipped real agents that handle real customer workflows in production. With a growing customer base and live deployments, we're scaling up fast and looking for deeply technical builders who want to have outsized impact early.

See more about the role & apply here: https://www.linkedin.com/jobs/view/4281413302

IMO this diagnosis is still one level away from a more fundamental truism, which is that people don't want to pay anything for digital goods. Running servers can and has been massively simplified over the last couple decades, and I don't see any inherent technical barrier preventing it from being as simple as registering for an account on FB (i.e. anyone can do it). The deeper problem is the lack of willingness to pay (directly) for anything online.

The reason for this is complex, with lots of unclear cause and effect dynamics (e.g. did our unwillingness to pay push the ecosystem to gravitate towards ad-based revenue models, or the other way around?). The inevitable race to the bottom between competitors, under the massive incentive for platforms to centralize/consolidate (if you charged any amount for your service I can always under-price and out-compete you) is likely a major contributor. We do not exhibit such reservations against payment for anything physical, probably because of the innate sense we have that anything in physical reality should have a cost, yet not so in the digital world.

Ready Education (YC S16) | Full Stack, Integration, Director | REMOTE | Fulltime | https://www.readyeducation.com

We are an education technology company, providing the leading mobile platform for universities and colleges across North America. We deeply care about student success, and work hard to make sure that students stay in school, stay engaged, and graduate successfully.

We are a distributed workforce, and fully embrace remote work. Currently we are hiring for multiple positions in remote roles across North America.

- Senior Full-stack Web Engineer: https://angel.co/company/oohlala-mobile/jobs/873383-senior-f...

- Senior Integration Engineer: https://angel.co/company/oohlala-mobile/jobs/873376-senior-i...

- Director of Engineering: https://angel.co/company/oohlala-mobile/jobs/881729-director...

Ready Education (YC S16) | Implementation Engineer (Java) | Montreal, QC | Fulltime, Onsite | https://www.readyeducation.com

We are an education technology company, providing a mobile platform for universities and colleges across North America. We deeply care about student success, and work hard to make sure that students stay in school, stay engaged, and graduate successfully.

Currently we are hiring for an implementation engineer position in Montreal. For more details including the job description: https://angel.co/company/oohlala-mobile/jobs/522261-integrat...

Forget a horse, the same argument can be made on a person's own self. Our conscious mental efforts cannot even "total takeover" our own bodies.

- This particular problem was artificially constructed with the specific goal of demonstrating quantum supremacy. It has little to none practical use. Essentially they looked what the easiest thing is for current quantum computers to do, and formulated a problem based on that

- Because of the above point, this does not mean that the quantum computer can do anything useful (definitely not factor numbers). It is nowhere near high fidelity enough to run highly error corrected algorithms (which Shor's and Grover's algorithms demand)

Since this is a mostly theoretical exercise to begin with, it's weird that they mention the computational cost of the classical method in terms of real computer times on supercomputers, and almost don't mention the theoretical bounds. The practical cost doesn't matter much as far as demonstrating quantum supremacy is concerned.

As far as I know, the key to claiming true quantum supremacy in this case, is actually the proof of the theoretical complexity of the classical algorithm. The quantum computer is obviously efficient at solving the problem, considering that the problem was constructed with what a QC is good at in mind. And the hardness of any classical algorithms had been somewhat demonstrated already, by Aaronson and Arkhipov in 2011. They managed to show that if there is a polynomial time classical algorithm capable of solving this sampling problem, then the polynomial hierarchy would collapse, which is seen as extremely unlikely (on the same level as showing P=NP).

PS: Aaronson recently gave a 3-part lecture at ETH as part of the annual Paul Bernays lectures. Links to recordings and PPTs here: https://www.scottaaronson.com/blog/?p=4301. Part 3 is specifically about this topic, and gives a good high level overview of the current state.

Edit: link to more info on the Aaronson and Arkhipov result, including link to the original paper: https://gilkalai.wordpress.com/2010/11/17/aaronson-and-arkhi...

Ready Education (YC S16) | Implementation Engineer (Java) | Montreal, QC | Fulltime, Onsite | https://www.readyeducation.com

We are an education technology company, providing a mobile platform for universities and colleges across North America. We deeply care about student success, and work hard to make sure that students stay in school, stay engaged, and graduate successfully.

Currently we are hiring for an implementation engineer position in Montreal. For more details including the job description: https://angel.co/oohlala-mobile/jobs/522261-implementation-e...

More generally speaking, when you optimize on a particular set of metrics to the extreme (when it comes to social issues this is what utopias try to do), you will inevitably cause another set of metrics to be correspondingly de-optimized to the extreme (which can be qualified as dystopian).

There is no way to optimize everything simultaneously because many things are fundamentally inversely correlated with each other (e.g. security vs freedom). So you either have a state that is relatively balanced (everything is mediocre), or a state with more spread (some aspects are really good and some are really bad).

There is a video (https://www.youtube.com/watch?v=S4zfmcTC5bM) from the PBS Infinite Series that covered this topic as well, for people that what a visual version of the explanation. I'm still sad that the series/channel was shutdown though, such a great series.

PS: this article/video essentially defines the naturals from the fundamental set theory axioms. This other video (https://www.youtube.com/watch?v=KTUVdXI2vng) from the same PBS series shows how you can then use the naturals to construct other types of common numbers, up to the reals.

Regardless of other reasons for this observation, I think sampling bias, or the inspection paradox probably contributes to it.

Assuming that you are:

- equally likely to have the desire to learn/create something new during any given time period

- busy for the majority of your time (e.g. you are employed in a time-consuming job)

Then you are more likely to find that you want to learn something new while you are busy (since whenever you randomly find something you want to learn, you are more likely to be busy than not).

If you swap the second assumption to "idle for the majority of your time (e.g. you are unemployed, or employed at a more relaxing job)", then you are probably more likely to make the opposite observation.

Last I heard, the original Big K, along with (some of) the secondary copies, will not be sold. Instead, they will be observed/studied in an on-going basis, in order to better understand why their masses appear to have drifted apart over the years (which is partially the reason for this re-definition).

I'm not sure that I understand what you are asking, since "prove that they have the insurance" seems to me to be equivalent to "telling the prospective employee".

Conceptually, I feel like there might be a connection to zero knowledge proofs (https://en.wikipedia.org/wiki/Zero-knowledge_proof), so perhaps you can look there for inspiration if you haven't heard of it.

Hi Scott,

This may be an ill-formed question, but it's something I've been thinking about for a long time:

Do you think the human mind is equivalent to Turing machines, or somehow above it? Assuming we have an infinite tape/memory and time.

The San Bernardino case involved an older device though, the 5C (which came out in 2013), which did not have the dedicated security module (Secure Enclave) at all.

As far as I know, there is no confirmation on whether Apple (or anyone) could flash new firmware to the Secure Enclave, without the user passcode, without wiping data on the phone. This info is strangely missing from the official iOS Security Guide document. If anyone has more info on this please share.

There are some (unsubstantiated IMO) claims by people online (e.g. https://blog.trailofbits.com/2016/02/17/apple-can-comply-wit...), and a series of Tweets with an ex-Apple security engineer (https://twitter.com/JohnHedge/status/699882614212075520), but nothing official. SEP firmware definitely can be upgraded without a key wipe (as confirmed by the Tweets as well as regular usage of iOS), but it's unsure if can be done without the user passcode. iOS does prompt the user for passcode when performing OS updates (which is also the delivery mechanism for Secure Enclave firmware upgrades). I don't know whether this is a UX-level security check only or actually hardware level required step.

Isn't this just that colleges are paying the market rate for good executives because they want to get talented and experienced people to run them well? Colleges are very complex and are definitely no easier to run than similarly sized companies.

SIDH (https://en.wikipedia.org/wiki/Supersingular_isogeny_key_exch...) is one of the few popular post-quantum variants of DH key exchange, and it supports forward secrecy as well.

Thanks for the info!

Did you guys write your own implementation or use an existing server?

This post comes at an interesting timing for us. Our product has a real time chat component, and currently is done over XMPP. We implemented using custom components on top of Tigase, a Java based XMPP server.

Due to a variety of reasons our chat services overall have fallen out of shape and we are in the process of considering a full re-write. Part of the consideration is ditching XMPP altogether. It seemed to me that for products where chat is only one of the features, XMPP feels like an awkward add-on, requiring its own set of IDs and auth protocols.

AFAIK many popular chat services these days are no longer based on XMPP, for one reason or another (e.g. scalability). Would like to hear any recommendations on what the best ways to do real time chat services nowadays.

OOHLALA Mobile (YC S16) | DevOps, Infrastructure Engineer | Montreal | https://oohlalamobile.com/

We build a mobile student engagement platform for universities and colleges. More details about company and position: https://angel.co/oohlala-mobile/jobs/261341-infrastructure-e...

Apply at tech.jobs@oohlalamobile.com.

I tend to think of "generalization", in the general sense (pun not intended), to be information compression.

We are simply trying to answer the question: what is the shortest description (i.e. most informationally compressed/dense version) that fits what we see in this infinite (at least to us mortals) universe of ours? Mathematically the length of such a description can be thought of as the Kolmogorov complexity.

Edit: I should add, the information compression performed when generalizing can (and often is) lossy.

the algo faq gave sort of an answer: https://github.com/trailofbits/algo/blob/master/docs/faq.md#...

thanks for the info. for the maze example, a ND robot would be able to find the exit in the optimal number of steps since one of the decision trees will in fact be the optimal path and will terminate first. so i dont see how analog computing can be more efficient.

unless you are saying that in an analog case we can ignore the rules of the maze and can go through walls. but then are we still solving the same problem? seems like the time complexity comparison no longer makes sense if you redefine the constraints of the problem.

You don't actually want qubits, you want an analog computer with differentiable signals. Most likely photonic. Qubits are a dead evolution branch.

that's quite a bold claim. im pretty intrigued, but it seems like there isn't a lot of easily accessible information on this topic. the wikipedia article on analog computers mostly focuses on mechanical ones from the past. do you have any recommended starter readings for analog computation? is it a model that has the computational time complexity of a non-deterministic turing machine? thanks.

Typically the forks include many performance tweaks over vanilla and so generally perform better. Here is an old test showing 5.5 differences in performance: http://vbtechsupport.com/657/2/

Percona's fork is also much closer to vanilla (than MariaDB), so there shouldn't be any issues even for existing databases. Unless maybe if the system/cluster set up is very complex so there is additional cost for migration. For new projects though, I can't think of any reason to use vanilla over one of the forks.

Is there any reason nowadays not to deploy one of the more performant forks of MySQL (e.g. Percona, MariaDB)?

i wonder what the complexity of the generalized version of the problem is. it is a bit similar to subset sum, which is np complete.

in this problem we have 4 operators (+, -, /, *) to consider instead of just 1 (+), so it is more complex than subset sum in this regard. also as a result of having different operators, the ordering of the numbers matter in this case, which doesn't matter in subset sum. however the fact that we have to use each number exactly once significantly reduces the search space

Will paid accounts get faster and more responsive service on the website? One of the biggest issues we have with GitLab right now is the speed and reliability of the website and service. Very often (for a service used/depended on daily by our team) the website would be extremely slow to load, and sometimes our CI/CD pipeline would hang.

I wonder if 1Password is equally susceptible or less so, due to the way that the extension works. Because 1Password has a native application, I believe the browser extensions merely communicate with the native application to retrieve passwords to fill when needed, instead of handling your whole decrypted vault.

Does this mean TB3 (with "only" 40Gbps bandwidth) won't be able to support all the features/resolutions of HDMI 2.1? This is of course assuming that the HDMI 2.1 spec is designed to fully utilize the hardware level bandwidth.

Also I wonder what exactly is making the 48G cable special. Is it using additional pins while maintaining backwards compatibility?