cjonas

11 ms·

211.
by cjonas 3y ago
+ What I'd really like to see in this space is "the supabase for business application development". Basically a framework like this, but with complete focus on developer experience. Would handle all the challenging infrastru
212.
by cjonas 3y ago
Congrats on launch. This is a space I've thought about a lot in my 12 years as CTO of a consulting agency that centers around Salesforce. Salesforce's "killer" feature really is the Custom Object. You can create a new d
213.
by cjonas 3y ago
To clarify even further: Jailbreaking is an attack against an LLM's "alignment"
214.
by cjonas 3y ago
If that scenario exists, is not a problem with the LLM, but with the fundamental application architecture... That's the equivalent of an API that allows the client to pass a user ID without auth check
215.
by cjonas 3y ago
Exactly... And if we properly design our systems to treat LLM output as "untrusted input" (similar to an http request coming from a client) then there is no real "security concerns" for systems that leverage LLM
216.
by cjonas 3y ago
In theory this works but typically in practice is not very effective because the context you add to bypass the prompt also impacts the effectiveness of the injected goal (since the entire prompt will get passed to the final eval).
217.
by cjonas 3y ago
Most use cases that actually require this much data are probably best solved by more traditional ML architectures (ie classification). LLM work best on use cases where the working context is the length of a short research paper (or less).
218.
by cjonas 3y ago
Interesting you put "eat" here as your first item. I don't care much about my rating (I play chess to escape stress; not create more of it), but over the last couple years since I started playing, my lichess rating (fluctuati
219.
by cjonas 3y ago
Who is the "solo" pricing targeting? I can't think of a single reason someone would pay $299 for "Non-commercial use" license to a library...
220.
by cjonas 3y ago
Ya, all it's doing is just adding some context to the system prompt to tune the way it responds. These custom "GPTs" really just feel like a honey pot to capture good AI ideas...
221.
by cjonas 3y ago
So in other words... Artificial intelligence? LLM are surprisingly effective as general AI. Tasks that used to require a full on ML team are now accessible with 10 minutes of "prompting".
222.
by cjonas 3y ago
The large context is the reason its hallucinating and gpt would likely have the same problem. You can see this effect even on gpt-3.5-turbo-16k. Performance and accuracy are directly correlated with context size
223.
by cjonas 3y ago
> I should generate UUIDs at every interaction and hope that collisions don't arise Assuming you're using uuidv4, the chances of a single collision is so low you don't need "hope"
224.
by cjonas 3y ago
The other problem with low code is that the visual abstraction can often make it harder to grok. I've had the unfortunate experience of spending hours trying to click through different nodes of a moderately complex Salesforce flow to
225.
by cjonas 3y ago
What's the workflow for leveraging this extension in real-time for an existing database? Say I wanted to use this to create a high performance "aggregation" API of my existing "write heavy" tables. Is there a way to
226.
by cjonas 3y ago
Except with many of these enterprise solutions (SAP, Netsuite, Salesforce) most companies will still need to pay consultants 300 dollars an hour to configure and maintain the application. These costs often exceed the license cost itself. B
227.
by cjonas 3y ago
Nice, awesome that you can just repoint an OpenAI client to this service. I'll give it a try. Do you have future plans to support fine-tuning via pay-per-token billing? Related: Found an interesting Reddit post (rare) discussing how O
228.
by cjonas 3y ago
I was just talking to a colleague about the need for a service like this. Their pricing is much cheaper than self-hosting on AWS (ml.g5.48xlarge, needed to run 70B Lamma2, is $20.36/hr). But the TOS makes it completely unviable. If s
229.
by cjonas 3y ago
Pretty sure this is based on one of those "login with Facebook" oAuth apps where your just using Facebook as the IDP. They were probably using the token to lookup the email address (via 'profile' scope) but not validatin
230.
by cjonas 3y ago
This is a real effect where newer (and maybe objectively better) tech is now at a disadvantage because LLM hasn't been trained on it (or there aren't enough examples to yield good results). Of course being an "early adopter&q
231.
by cjonas 3y ago
> We do have plans to improve the current database trigger behavior. Will share more updates on this in the coming months. Excited to hear more! Being able to trigger functions from DB & cron triggers without having to hardcode a se
232.
by cjonas 3y ago
100% agree... I've been using super base for a new project for a few months and while overall I'm impressed, there are just too many bugs and partially rolled out features. Here's my current gripe list: - features/ui mi
233.
by cjonas 3y ago
In the keyboard event example of only specifying the properties you want to test against, it would be better to use `Pick` to create a relational type off the original. That way if the property or types change (unlikely in this case), your
234.
by cjonas 3y ago
In all serious, I've been talking with my colleague a lot about this. Will be interesting to see if mining archives of "low background steel" data (anything created before chatGPT exploded) will become a industry. Another int
235.
by cjonas 3y ago
In the future, we will have air gapped facilities where people are paid to harvest their thoughts, similar to plasma "donation" ;)
236.
by cjonas 3y ago
Maybe trivial isn't the right word, but it's still very straight-forward to get something basic, yet really powerful... ReAct Setup Prompt (goal + available actions) -> Agent "ReAction" -> Parse & Execute Actio
237.
by cjonas 3y ago
A lot of software is dependant on servers so it somewhat understandable (at least for the usecases where the server is essential)
238.
by cjonas 3y ago
I've had the same experience and gripes as the comment op, but RTKquery is one of the better experiences I've had with OpenApi based tooling. But ya... JSONschema is confusing and doesn't really support type composition the w
239.
by cjonas 3y ago
In my testing with GPT-4 (via API), it handles this fine if I split the prompt into multiple user inputs and provide clear instruction (provide a summary and nothing else). But ya, there are an infinite way to attack this.
240.
by cjonas 3y ago
GPT-4 seems to handle it ok in my testing... but there are infinite variations of this, and I'm sure some of them would work.

More ›