jerrygenser

Google puts things on the chopping block due to small userbase or not enough user growth. A very large user base product but which has stagnated should be fine as long as users don't start leavign

It's a lot of electricity to power it and heat that gets generated... and electricity to cool it from the heat

> Isn’t that very difficult to do? Resolvers have recursive relations that can be very tricky to secure, since it’s a high abstraction level. So you tend to rely on RLS or another system to be configured, no?

I really don't think it's any more difficult than if you're not using graphql. If you're hitting a db then you can store auth/permission context for the lifetime of the request. Each time you hit the db to resolve a node, you're going to use the same types of patterns you'd use when fetching objects from the db using sql or an orm. It's customary to have a request context object available in all of the resolvers for just this reason.

If you resolving by calling other services, auth scopes would typically be forwarded on in request headers.

> All popular frameworks I’m aware of are selling themselves as having automatic db resolvers with SQL dbs.

I think I agree if some frameworks for full stack aimed at prototyping where a solo frontend-focused developer needs a backend it abstracts too much, including the db. In these cases, it's probably unnecessary and if you're solo dev on a full stack framework and probably doesn't matter either way.

I've worked on graphql at scale as an frontend service that acted as an API gateway of sorts. The resolvers were not accessing the db but calling other services.

When working on a new application, I would probably use REST and probably would be a long while and would need to have a set of problems to solve with graphql. I think it's quite good at federating services.

It's perfectly fine to use graphql for fetching metadata and then using REST API for file upload and downloads.

On a gql server, you implement resolvers for the schema. There might be some frameworks that abstract SQL away from the resolver implementation but a resolver may need to make a combination of sql and network calls to other services. Or only network calls to other services.

> Allowing frontend to query db directly is insecure so you must have fool proof row level security or more resolver complexity. I'm not sure why you are thinking that graphql means the frontend is querying the db directly; a graphql server implementation is implement a callback for each graph node that returns the object. It's the server code logic to implement permission checks and the actual SQL or network calls to resolve the object. What part of this results in frontend code querying SQL directly?

It seems like you are getting caught up in the name of "query language" which may be partially a misnomer. It's more like a data fetching protocol

I wonder what the rationale behind AGPL is? Prevents it from being used for commercial SaaS

I think it's clearly better to have the email eventually sent than the case where a completely synchronous service is down resulting in no email being sent

Is this because of the elastic search license or simply the project author chose to commercialize this functionality?

Intro video is private. I could not access it on YouTube.

https://openobserve.ai/docs/getting-started/

Django ninja performs better primarily due to relying on pydantic for parsing and validating requests. In a high throughput scenario that's going to be the primary CPU bound area. However usually network or file io is going to be a limiting factor far before CPU.

Django ninja also is build with async in mind fron the start so immediately opens up possibilities that are not possible with DRF.

Seems like as of 4mo ago the maintainers of DRF added async support via a separate module.

Or just use django-ninja if you are writing an API. Maybe it's just because I came from teams that used tornado and then fastapi but when I started using django which was about 8 months ago, I simply grabbed django-ninja and haven't thought much about it. It seems like everything in this article would be solved by using a simpler interface for writing endpoints with body parsing, output schemas, and url parsing.

https://django-ninja.rest-framework.com/

My 49 inch 144hz monitor ran fine for this exact use case when I had an amd 3600 and Nvidia 1660. Is your work use case somehow more demanding than gaming?

In general both on my own benchmarks and from what I've read/seen youtube videos, it's about a 5% performance loss multithreaded with full core saturation and 0-1% performance haircut on single threaded -- since on single threaded you are generating lower enough heat not to go near the high end.

For me this was going from 105W to 65W on my 7700X and my results were similar.

Not sure how it would look or work with Intel though.

For your gaming PC you likely don't even need a current gen chip. You are probably fine with a 5600 for most games unless you NEED high FPS on certain games and have a monitor with 244hz refresh rate (or you won't even get the frames) Gaming is mostly about GPU (if you're casual gamer and want to run 1440p) but if you do need cpu and high refresh rate then you're likely hardcore FPS or RTS games and you're running at 1080P anyway and then in that situation don't need a high end GPU.

I have a AMD 7700X and I run it on Eco Mode which is approx 65W TDP instead of the 105W TDP it wants to run at.

For my general use (including cpu intensive operations) this makes absolutely no perceivable difference. I ran some benchmarks out of curiosity and I take about a 5% haircut for a massive power and heat savings.

What Intel and AMD did with the chips was essentially sell consumers a default overclocked chip that will run to the max of the thermal headroom that your fan will allow and sit there. They did this to be competitive with each other in benchmarks for marketing.

Most consumers should run these chips in some sort of eco mode since the performance per watt has severe diminishing returns and they are actually quite power efficient as long as you don't run it on the default factory overclocked settings.

> Having said that, the 7950X was released late February, and the 13900KS was released in mid January. Both of this year. Both are their premiere available chips right now in the segment. Referring to them like they're last year's junk is rather silly.

7950X was released September, 2022. It's quite literally last year's chip and given that AMD release cycle is typically about 2 years, we're roughly halfway between last release and 9000 series release.

You might be getting it confused with the Non-X versions that were released earlier in 2023 -- those are basically the same chip but power limited and maybe slightly worse selections of silicon. Of those Non-X versions it was 7600, 7700, 7900 but non-X of 7950 was released. [1]

[1] https://en.wikipedia.org/wiki/List_of_AMD_Ryzen_processors

errata - that helm templating used Jinja. It uses gotemplate

It seems to be published as research not as a product announcement.

It's not absurd. It's because these tax prep companies spend a lot of money lobbying and putting those congresspeople in office.

Does this take into account failure to consider context? I have a lot of trouble with bard not using context from previous messages.

What about if you were to use it to write code but the code itself has logic to restrict what it could do based on the execution environment. Whether it's external variables like email Allowlist or flagging emails as not allowed. Your assistant if it tried, it would not have access.

In that sense I agree it could be a problem solved without "ai". Simon's approach does use another language model, maybe we need to build more way of logically sandboxing code or just better fine grained access control

Paying for chatgpt I believe is separate from API access

[1] is quoting a Twitter account that is quoting a parody Twitter account. I don't believe that is a thing he actually tweeted or said.

This is a nitpick but also it's been a few years since I was taking academic ML and linear algebra courses. However regarding this part.

> So you only gain a factor 2 * L. But the trade-off is that you are doing a near neighbor search in higher dimension where you are then subjected to the curse of dimensionality (the higher the dimension the more similar all points are to each other).

I thought that the curse of dimensionality meant that in higher dimension, points got farther apart

Would using llama as the llm teacher for a commercial model mitigate licensing concerns since it's not being used directly but only to create a separate artifacts?

What could impact the ability to discern whether it's llama or open llama? Are there artifacts of a given Llm that could be exposed via distilling process?

They haven't trained a 1 trillion token model yet. They have only done 200bn so far

Sorry this isn't the purpose of the original article, but does anyone want to give a way secrete should be stored? Even if I use secrets manager for example, I need to store the token for fetching it so there is always a secret or token that needs to be stored.

I might not store an access token, I might use user auth and get a temporary token for a script that then calls to a secret manager. The access token is cached locally in plain text. Is tbis more secure since it's a token with expiry?

I was interested in looking up sqlc for python but it seems to be for go? Is there a different version than this? https://github.com/kyleconroy/sqlc

Is creativity not a mashup in and of itself? A human brain learning a bunch and then combining what it knows throwing some randomness in there to create something... "New"

In pure scientific reasoning I agree the models need to be able to encode logic somehow.

In any event, by mashing things up you might concede that these models can at least encourage human creativity

Agreed, it's kind of really useful for a lot of things people have been using NLP to solve. In some ways the human interaction piece expand the scope but overall it doesn't mean NLP suddenly solved all problems

Not loki but regarding tempo, I've had good experience with running the components in kubernetes using the operator.

These components are typically build to be "cloud-native" which often means run on kubernetes. If you already run on kunernetes, grafana products are typically straightforward to run.

In general I think that is the target customer and use case for out of grafana cloud deployments. The all in one binaries are more toys to say hello world or maybe test things in local development.

I could definitely be wrong but I had it answer me prompts like below (all are verbatim), which led me to this deduction but perhaps these are hallucinations:

(Forgot the exact prompt that led to this)

> I am FastGPT, an AI assistant built by Kagi Search. I was not trained by any organization. I am an AI model created by Anthropic, PBC to be helpful, harmless, and honest using a technique called Constitutional AI.

"Who created you?"

> I was created by Anthropic, PBC, an AI safety startup based in San Francisco.

"Who trained your model?"

> I was trained by researchers at Anthropic, PBC, an AI safety startup based in San Francisco.