Jorge Colon

Starting the day at the gym for another productive day. Yesterday client was happy with AI Coding tutoring on Claude Code w/ the BMAD Method and Cursor. Liking it so much he’s booking 3rd session. If you have more money than time and want a tutoring session, book a session or connect with me to see if it’s a good fit for you. calendly.com/jorgecoloncons…

@0xDesigner It’d be nice to pair this with Gemini 3.5 Flash feeding it the video to build the context since it has native support for understanding video, not just images. Read it’s really good at it and cheap.

WOW 🤯🤯🤯 i made a 15 min screen recording and talked through every UI bug i could find. pasted the screen recording in claude code and just said "workflow, go" much better result than a text prompt or asking it to verify with computer. highly recommend trying this if you haven't already.

@RafaelNegronX @boshen_c Hit me up and I’ll put you in the right direction

@boshen_c I’d love to learn about what specific workflows you used during your work. I feel like I’m still mainly using AI as a chatbot, and I need to level up. Any resources you can point me to or experiences of your own? Thanks!

I spent $6k worth of tokens in the last 3 months but why do I feel so empty inside. Did I do meaningful or impactful work? Did I actually learn anything? Did my engineering skills improve? Did my peers liked my slop PRs? Nope!

@iruletheworldmo Did a large feature/refactor after nailing the details with superpowers using Opus 4.8 xhigh in auto mode. It went on for over an hour going through each task and following instructions closely. Code quality was high, but I attribute it more to my workflow than just Opus 4.8.

@iruletheworldmo I’m liking it so far, but it’s hard to tell if it’s a massive step up because I have a very specific workflow. It pretty much works regardless of new model releases. I don’t use Opus alone and I actually have codex 5.5 high do adversarial reviews to catch what Opus 4.8 misses.

so what’s the vibe for opus 4.8 vs chatgpt 5.5? seems like there’s no need to switch back to opus? roon has the mandate still?

@NeilSanghrajka @claudeai Technically Meta+w should ignore it but looks like there’s a bug and it still attempts to create the dynamic workflow

so now I can't use the word "workflow" in my prompt, without @claudeai automatically enabling dynamic workflows (costs a lot of tokens)

Today, we're releasing LFM2.5-8B-A1B, a device-optimized model designed to power real-life applications on phones, laptops, PCs, robots, and fast & lightweight server-side use-cases. > 8B MoE, 1.5B active > Expanded 128K context > LFM2.5 flagship hybrid MoE architecture > Trained on 38T tokens + large-scale RL > fast, reliable tool calling, punching above its weight, comparable to models with up to 4x its size > customizable on a single GPU for any specialized task > LFM2 open-weight license 🧵

@Metropoles Não tem problema que brasileiro não resolve com gambiarra

🚙😮 Mecânico criativo cria gambiarra para ajudar no trabalho e ensina os seguidores: "Estou patenteando" 🤳 santiago.mec01 | dommedia.br

This is why it’s important to actually have an accurate understanding of how much a task costs to complete. Just gauging how expensive or cheap a model is based on input and output prices isn’t enough. If a cheap model burns a lot of tokens for the same task or expensive model uses very little tokens for the same task then that should be taken into account as well.

💯 this is why I really like Learning mode in Claude Code I personally use this for all my side projects and it keeps me so much sharper, great if you want to use Claude Code but still stay hands-on! /config → Output style → Learning

@girdley Zight

Has anyone found a Loom alternative they love? It's so buggy now that Atlassian is behind it. I'm sick of recording 5 minute videos that it loses.

We are going to sue YouTube But before we do, our lawyer is currently in the process of sending a letter to YouTube. Over 2 weeks ago our channel @UnderSparked" target="_blank" rel="nofollow noopener">youtube.com/@UnderSparked was falsely demonetized under the "inauthentic content" policy. Demonetized doesn't mean no ads though, all it means is YouTube will still run ads on the videos. But they pocket all the revenue, essentially stripping all the revenue from us. We don't claim to be the pinnacle of quality content standards, but the following things do cause us to scratch our heads in confusion 1. All of the bigger channels in the same niche as us are monetized with no issues 2. We literally voice act, edit, etc, all our own videos while 100s of AI storytelling channels are still monetized 3. YouTube is fine with running ads on the videos, pocketing all the revenue. We assume if YouTube had such a big problem with the content, why still serve it to your advertisers? 4. 100s of reaction channels still stay monetized Some examples of bigger monetized channels that are in the same niche as us are: rSlash: @rSlash" target="_blank" rel="nofollow noopener">youtube.com/@rSlash DarkFluff: @DarkFluff" target="_blank" rel="nofollow noopener">youtube.com/@DarkFluff Oz's Vault: youtube.com/channel/UCoJ1A… We support these channels, the reason we are providing these channels as examples is to showcase how it feels like we're being targeted. This has caused extreme distress for my team and I. Many of our livelihoods and families rely on this channel. Our community relies on the UnderSparked channel. Many people support the UnderSparked channel because in a time of nonstop AI channels, we stepped up and used real voice actors to cater to that connection and community that viewers on YouTube crave. We truly believe this demonetization was done in error, all we want is a real human specialist to help us fix the issue. We're tired of constantly receiving only copy and pasted templated responses, constantly neglecting/dismissing our attempts at fixing the false demonetization. We're kindly requesting for a representative from @TeamYouTube to help us fix this demonetization issue with our channel. We don't want to send letters, or sue, or anything! We simply just want to go back to producing content for our community.

@TansuYegen I’m thinking huge liability risk

Billboards met drones, ads suddenly moved into the sky above cities 🚁

“Why change” really depends on if you’re repeating the same thing over and over and it’s getting annoying to you. Especially being overly verbose about the same action and the model tripping up a lot on the same thing. Perfect candidate for a skill. For instance, I have a local codex action in Claude Code so I could have GPT models review or sometimes implement Opus’s plan. Without the skill it makes a lot of mistakes. With the skill, it becomes a lot more deterministic. I also use the caveman skill to save on tokens, make responses more terse, and faster. Superpowers skill to help build thorough plans for complex features based on TDD. I turn on a couple of MCPs as necessary depending on the task at hand.

Feel like I'm missing out because I don't use skills, or a lot of MCP, or multi-agent orchestrations when using AI dev tools. I'm just like "implement this feature" or "how do this work" or "no not like that, do this instead". Idk, I feel fast and accurate so why change?

Introducing SubQ - a major breakthrough in LLM intelligence. It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA), And the first frontier model with a 12 million token context window which is: - 52x faster than FlashAttention at 1MM tokens - Less than 5% the cost of Opus Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention). Only a small fraction actually matter. @subquadratic finds and focuses only on the ones that do. That's nearly 1,000x less compute and a new way for LLMs to scale.

@oprydai Did he mention what model(s) he uses?

met a doctor this morning at my university, completely from a clinical background; who told me that he has setup a GPU rig at home with an RTX PRO 6000; and is hosting LLMs for his family. Also built an app using Codex that he and his kids use. his main concern was privacy, "I just don't wanna share mine or my kid's conversations with these big companies, as a dad this is my responsibility to educate and protect my family." honestly, I have genuine respect for people like these.

Everyone needs to realize we haven't been writing most of the code for *decades* now, thanks to frameworks, libraries, IDEs, intellisense etc. This is the same, just faster, more code, and a little more reckless

Also, experienced devs can make AI agents write clean, maintainable code just as quickly as a junior dev can prompt crappy code. No excuses.

This. There’s times where parallel agents make sense, but if your workflow speed runs massive amounts of AI slop you need to reevaluate your process.

@HumanityChad Havaianas te dá aaasas

Quick thinking man ran out of a nearby shop to control a runaway car with kids inside

FINALLY