1000 results for “ll”
Llevaba un tiempo sin jugar a nada por falta de tiempo pero como fan de la saga le tenía muchas ganas al nuevo Castlevania, y por ahora no está decepcionado.
Ese regusto a clásico con una jugabilidad cuidada y un arte que me está gustando mucho, todo por 29€, estando como está la industria pocas alegrías como estas tenemos últimamente #gaming
LLMs are a prism, splitting coders into two categories which were previously impossible to separate: those who code for the craft and those who want to get things done. Which one are you?
I work on the APIGOTO team. APIGOTO is independent and not affiliated with OpenAI, Anthropic or Google.
APIGOTO is an OpenAI-compatible LLM gateway: one base_url https://api.apigoto.com/v1 for Claude, GPT and Gemini, automatic failover capped at 3 retries, per-request logs, BYOK, pay-per-use.
Question: is a fixed cap of 3 retries a sensible default, or should it be configurable per request?
www.china-aiapi.com
#LLM #API #DevTools
Watch “Agents Need Lift Stickers, Too”🛗
Verification is more convenient than blind trust, always.✅
#LLAP🖖
RE: https://bsky.app/profile/did:plc:myh6qcwcvg53u3nqsvzkljq6/post/3mxwbwzwd6s2s
Llegó la hora de comer y oficialmente abro el debate del día: pasarse 30 minutos buscando qué ver en la TV mientras la comida se enfría por completo. Es un clásico de la vida moderna. A ver con qué joya o desastre me topo hoy en el zapping para acompañar mi platillo.
¿Alguien más sufre con este dilema diario o ya tienen su serie fija del momento?
Alt text via @altbot y @TeLoDescribot
#televisión #horadecomer #Zapping #series
(FYI: this is going to be a thread discussing practical AI-coding problems, not the deeper problems. If you want to argue about never using AI, please don't do so here.)
#LLVM has a proposal to ban using AI to draft PR descriptions and comments (and some other things), no matter how much human-in-the-loop. I've felt pretty good at getting (with sadly large amounts of effort) these tools to write first drafts for me, and I shared that on the LLVM discussion thread[1]. Now I'm working on a follow-up, and brainstorming here first -- comments and discussion welcome!
After some private push back that I was wrong about how even my current efforts were working (thanks BTW!!), I tried something, and I think I'm changing my mind.
First, I spent some fun coding time using these tools, but switching to always drafting PR descriptions and comments myself. I also had the first drafts for a some of these PRs already written and I _thought_ edited into a better form from quite some time ago, so I could compare the results.
Two observations that surprised me:
1) They were much more different than I expected.
2) The differences weren't the kind of differences I expected.
(continued below, even my word limit isn't high enough)
Llevo 3 días intentando ayudar a una vecina a inscribirse en una oferta de empleo público de la junta de Andalucía pero es imposible, para subir los documentos se carga un Applet de java para firmar el documento con certificado digital y peta... He probado en Chrome, firefox, brave, Vivaldi...con instalaciones limpias, ultra permisivas, he probado normal e incógnito, con VPN y sin VPN y no hay manera... La cosa es que en el validador de certificado digital de la fmnt, agencia tributaria y web de nuestro ayuntamiento si que funciona bien el certificado digital y se valida y firma correctamente...pero la puñetera web del empleado público no.
Los admins me dicen que pruebe con VALiDe....una herramienta que sigue dependiendo de applets de java que nadie usa desde 2017.
La paradoja del "sigilo XXI" administrativo: sistemas digitales basados en tecnología obsoleta que solo funcionan por suerte y reinicios mágicos.
@CStamp Might I add an observation?
This is not merely a "predictable event".
This is the very purpose for what LLMs were created. They don't "steal IP and other people's work, to allow access to their value, for sole benefit of billionaire techbros", they... "SCRAPE".
Give it a cute name, then it's aaaaall better. No liability!
This is the fundamental purpose and M.O. of LLMs, because it is what they were designed to do. System is working as intended.
Llevar contabilidad no es lo mismo que enviarla al SAT. Entender esa diferencia puede ahorrarte más de un problema fiscal. 📋
#ContabilidadElectrónica #ObligacionesFiscales #SAT #ContabilidadMéxico
You may be missing the point.
They are panicked, because incident was known to outsiders, and they need to sanitize this, so their further exploits are not a liability to them.
So they push the anthropomorphic narrative that "the system" did X, so avoid responsibility. The LLM is NOT. AWARE.
The builders and users are liable for what the use of a tool causes.
LLM-based AppSec vendors are converging on features, so how are customers supposed to differentiate between them?
The difference is ultimately the offensive security expertise encoded into the core of the product, and the quality and relevance of the findings that come out of it.
We have world-class hackers who have spent years finding in-the-wild vulnerabilities in real-world software and systems, as well as working directly with customers through offensive security engagements. They have accumulated practical knowledge about where meaningful vulnerabilities tend to hide, how real attackers approach complex systems, and how to distinguish an interesting weakness from something that is actually exploitable.
That expertise — and how we harness it — is fundamental to Xint. Some of these researchers are directly involved in developing, experimenting with, and improving the core engine itself, so the knowledge they gained from finding real-world vulnerabilities is continuously being translated into the product.
How AI chatbots may be talking people out of climate action
Across 28,800 prompts, models were 7.5 times more likely to advise “stay the course” than “change,” even when the status quo is anti-climate.
I’m legit just starting to grasp how immensely powerful the machines that we have today are.
Once you strip away the layers upon layers of complexity we accept / fetishize / cherish, there is no going back.
Concrete examples soon, but basically I’m now just using html files and tiny JavaScript for everything. I see friends porting apps to bare metal and booting in two seconds.
It shouldn’t really be shocking per se as someone who has done embedded for 20 years, but it actually is. It’s one thing seeing a microcontroller being snappy. It’s another one to be in disbelief because your apps start instantly and taking mere kilobytes.
Concrete example: I replaced my photo printing software with a tiny layout html + color profile adjustment. I replaced my cnc controller with a 500 kB html. I replaced my CAM software (!) with said html too. I replaced my bike trainer with some random html I change up whenever I feel like it. Apps on my embedded device are usually a couple dozen lines of JavaScript.
Why did I never do it by hand before since it’s so small? Because I had to rewire my brain to see everything as an opportunity to be replaced/reinvented, and because generating said html takes less time than fighting with the stupid nonsense we have.
Two examples I managed to put online:
Https://cam.yolo.scapegoat.dev because this took less time than figuring out how import a fucking jpg into my CAM, and has better algorithms.
Https://cam-ide.yolo.scapegoat.dev because well CAM sucks.
(These are old, I need to figure out a better way to share my stuff, I’m not great at releasing / publishing my stuff)
I think that using #llms for software can be moral, and in fact politically progressive.
Here’s why, so I can point to this as tldr.
- llms are revolutionary for software building.
- we can reinvent what software is: something everybody should be able to do in order to control what their computational devices do. That does not mean becoming an engineer, that means that generating code is just part of the runtime, of the operating system, if you want.
- we can reverse engineer every piece of obsoleted firmware out there, every piece of surveillance capitalism iot gadget, turning anything produced in the last 20 years into the perfectly usable devices they are. We can create operating systems that are geared towards power efficiency and simplicity. We can replace cloud apps with 200kB HTML files. If you haven’t experienced what that looks like, I can finally point at plenty of examples.
- whatever the fuck the industry is doing with tokenmaxxing should be countered by thoughtful exploration of the technology. Even the industry’s “let’s replicate the bulldozer approach” is at this point effective at creating something valuable enough that the industry is _never_ going to come back. I know for a fact you can create highly usable software with a handful of tokens.
- we have to face, as programmers, massive labor and maybe more importantly identity crisis. They cannot be ignored. We will have to create new labor models and organize, and yes if you want to continue producing software, that means learning how to use llms. I can do the work of a year or more an entire team in a few days or even hours with the latest models. It’s not about me enjoying it or not or wanting to be more productive, it’s a fact.
- we now have plenty of alternatives to OpenAI/anthropic models. We need more. Hyper scaling is by far not the only modality. We can’t let companies that move fast and break things be the only framing. We cannot leave the field to them. And yes that’s my framing of environmental concerns. The industry is not going to care, but we can continue working on understanding how to maximize the value for watt (see rethinking operating systems and reviving every piece of hardware out there).
Most universities treat support for deaf and hard-of-hearing students as something you request after you arrive. RIT built it into the foundation instead.
Every class at RIT, across every major, has interpreters available, so a deaf student can study mechanical engineering or computer science and get the same options as anyone else, with captioning and note-taking already there.
I wonder if the difficulty (or impossibility, with secure boot locked down chips) of replacing your bios / boot chain from start to end will become just a standard selling point. I was gonna get a t490 thinkpad to follow @hanshuebner ‘s experiments, but will revert to a t480 after he ran into locked fuses.
There should be a law anyway that companies should release whatever signing keys they used for older hardware, including all engineer documentation. I haven’t followed right to repair efforts / results in Europe (well guess I should just read up on it huh: https://repair.eu/news/the-state-of-right-to-repair-in-2026/)
Now that it’s a matter of an hour or two to replace your boot chain, this becomes a real differentiator.
LLM advocates love to claim constantly that LLMs produce better code than they or anyone else could be expected to write, and then you see the code and it's a hideous pile of garbage.
FYI: Explaining LLM: An LLM, or large language model, is a neural network trained on vast text to predict the next token, the engine behind ChatGPT, Gemini, Claude and AI search. https://ppc.land/llm/ #LLM #ArtificialIntelligence #MachineLearning #NaturalLanguageProcessing #ChatGPT
#LliureDirecte #Ara ja en #directe amb les #entrevistes i la vostra #participació LLIURE DIRECTE https://lliuredirecte.info - RÀDIO KLARA 104.4FM #València https://radioklara.org
LLMs are machines that turn software engineers into gambling addicts.
Just found this excellent post on how LLMs (maybe) change how we use language: https://lucumr.pocoo.org/2026/5/4/content-for-contents-sake/
I'm surprised “quietly” does not appear in the list of words with inflated usage, but the post is from May 2026, so perhaps it appeared on the scene later.