Some devil's advocate- AWS receives many support requests that initially go to customer support engineers (in call center environments) who act as first defense for pointless simple issues. But you still have ones escalated to actual engineers and technical teams, costing hundreds or thousands of dollars. There are times when its like actual service issues and bugs etc but the number of times you have a long thread between a support engineer who does not even understand what the customer is misunderstanding (because of arrogance like this, and the customers continuing assertion they are not wrong) can go on for days/weeks and is just monumental waste of time when the answer is read the docs.
MCP is eating whatever the next buzzword is until something else or better comes along. The hype cycles with these ancillary things (RAG etc) is so overblown compared to what actually makes it into practice and production. I would use MCPs as developer. In making AI systems, unless you are in large org (also rare) not even worth doing.
what? you have a multitenant system with isolation between users not within a user. OpenAI has literally been working on "memory" and context across everything you have done, they obviously are storing and retrieving recent context. I see no indication in this at all that they are using browser or local storage for anything, because we know they run other models to summarize and extract info from context. Likely you could turn it off if thats configurable yet.
This doesn't tell me anything about how you actually solve the problem in a value-add way. There's no technological insight? Are you just putting this behind an API (adding more latency). The thing about memory is for non academic cases the value of it is still nebulous and hard to quantify, another reason why it still feels experimental, and why I would find it hard sell to push that to a black box service. There's nothing on your website about actual performance...
talented and smart folks for sure but can't not notice how much luck it is especially because its like 100% just better models. Windsurf raised a ton of money and then said they pivoted which they had millions raised to just do something completely different that likely wouldn't have been easier to raise for. Even in an interview with the cursor founder he kind of dumbly rambles that they launched and then basically lost a ton of traction until GPT4 came out. They have some core features like autocomplete but I'm struggling to see vision other than getting training data for iterative dev is a partial moat compared to just seeing commits and final code bases.
idiotic comment. This is literally a training mission - not following an order is not disobeying this is clearly failure of instructor who was not ready to take over. This is a common pattern in like almost any other training situation.
Not really thinking discretionary goods are going to see a boon from tariffs. People buy a lot of things primarily because they are cheap/affordable which is why outsourcing and global/trade has benefits for consumers. There's a lot of stuff people will just go without, like overpriced backyard stoves....
the guiding promise of AI was able to figure out unstructured data. Manually crafting prompts and description examples for tools is a step orthogonal to progress. Searching for docs, great now you have 1000 tools saying they do same thing exposed to model, now we need another search on top of that?
shhhh - that's the big announcement for next year - the "Agnostic" General Interface (AGI) ... works with any endpoint no more lengthy MCP or A2A setup!
This is an incredibly poor take - curious as to how you just completely made up that ridiculous claim? Ironic as well that you think letters of recommendation matter for college admissions when they are perfunctory for probably > 95% of them. Maybe you shouldn't espouse your opinions on this.
um - I would. Colleges don't want to get panned for hiring a scam artist which is exactly what this is. Unlike VC where that skill is slightly revered.
The app is fake - at best its puffery, and the essay was littered with grammatical errors.
The over engineering here is commendable for something that provides most likely marginal value whilst still needing more api calls to summarize everything you do.
Find it disappointing that even on HN people seems to blindly parrot manufactured outrage. Do you really think, nay have you ever tried to look a little deeper than surface to see just how preposterous all around the idea is in terms of actual intents and relative numbers that it would be meaningful to "cut" them? US Republicans are not about states rights, its just the excuse they use when they are trying to dismantle or challenge systems.
I would not define ChatGPT as a wrapper, anything where you are doing actual training/learning and updating weights is by most persons definition, not a wrapper. Just injecting stuff into the context or using RAG is a wrapper because there are no weight updates anywhere.
kinda surprising to see basically just LLM wrappers still getting funded/interest. Neat idea obviously and nice to have, but that's sort of it. Fundamentally misses mark on what you can do already without AI (running tests on your actual doc code examples etc), and fact that any of the AI IDEs could turn part of this one with single switch to just append a doc friendly update to a commit. To think that soon we have ai agents reading ai generated docs instead of code to reduce it back down to code...
Funny enough this is almost a canonical example of data leakage and the incorrect (or extremely extrapolated) assertions that AI can solve what seems like majorly important and valuable tasks. There are 2 very pertinent observations needed.
1. The state of the healthcare system, assuming this is the US, care is based on triage and constantly requires a number of hoops to go through to get deeper analysis or specialists, an emergent property of not having value based care. They do this to save costs while ensuring you are not going to die.
2. They are using and feeding in data from the doctors visits, which means that AI is not really providing a diagnosis in a way that can replace anything necessarily. The fact there are many visits would help a flow chart decision rule out/in possibilities.
You see a lot of this in AI hoopla and even in some of the industry papers where AI is relying on data that only becomes available because of a doctor's orders, so you are not "replacing" it per se.
If anything this demonstrates that value based care would provide better experience and outcomes for people with hard to diagnose conditions.
A great litmus test for BS is when you find more articles written in the past week about something that happened 10+ years ago and all the armchair experts come out of the woodwork. This issue was basically asked and answered, and in the time sense no one has done anything meaningful to fix the structural problems which is almost entirely that they do not pay enough for a job that is fairly life consuming. They cannot even get enough applicants now, so the unfairness in selection would be moot as it does not exist. We just saw past year a bill struck down to increase pay, and I think a few weeks ago in California controllers quit because they restructured contract to not pay COL.
Those are facts that are pertinent, and we should strive to use more critical thinking when analyzing situations.
There was a truck that crashed on bridge right next to lincoln memorial and then veered off roadway into bridge, and both were deceased. Its very cold in DC right now and an airplane crashing is much more catastrophic.
you're looking at it from economic theory not from stock market. NVIDIA's insane valuation right now was based on an almost exponential increase in demand for more and more of it. It's priced in that NVIDIA will continue that trend. DeepSeek proves that trajectory is no longer needed (not that it was ever cemented in rationalism), so anything less than the continued exponential growth would send stock down.
every single model does/did this. Initially fine tuning required the expensive hand labeled outputs for RLHF. Generating your training data from that inherently encodes the learned distributions and improves performance, hence why some models would call themselves chatgpt despite not being openai models.
That seems like nonsense. Hotel chains and esp business travel ones it'svery standard for hotels to have desks and a chair. Not saying it's that nice but those are way cheaper than airbnbs usually
Abhorrently bad take and misses the biggest reason that documentation usually is a second class citizen, and that is because your code is law, and code and tests are coupled, while documentation is not (thats why there is such huge push for autogeneration). Even code comments are considered a code smell because they are not part of CICD process. The way they describe documentation here reads more like a feature story and not being able to fully complete a spec, which is a tangential problem.
For LLM aspect - My take is the better approach is similar context enrichment for rag - you ask the LLM to generate a summary for a page, and that can easily be cached against a digest/hash for any code changes - and will always be up to date.