If you know you've made a mistake, always say sorry no matter how small. You do not want to develop grudges in people who could have been your allies.

Everything is p*** in the new social media. Stay away from AI p***, politics p***, aesthetics p***, fomo p***, MAGA p***, wannabe rich p***. Weaponized content is to enrich big tech; they don't care about your growth and prosperity; they just want to hack your dopamine with fear.

What do you want to know?.

Scamsara.

Never ever idolize your heroes. They are all flawed humans. Instead, steal their best qualities and do not cultivate their worst.

RT @LeilaHormozi: One of the most underrated skills you can learn is the ability to ignore your mood and stick with the plan.

Amazon UX is just the worst.

We built a geocoder for the UAE from scratch because addresses here have no postal codes. No off-the-shelf tool worked. So we made one. It now powers routing, sorting, and ETAs. Sometimes the most impactful AI isn’t flashy. It’s the thing that makes the most impact to your.

One of the quietest AI wins I’ve been a part of?.Using AI to turn compliance into speed at a major logistics firm. We trained a model on IATA/UN rules to auto-flag dangerous goods before they hit the tarmac. → fewer errors.→ faster clearance.→ fewer fines. AI doesn't need to.

Hallucinations aren’t just “the model made up facts.” There are two types:. - Intrinsic: model fabricates based on training data (e.g., Webb Space Telescope exoplanet myth). - Extrinsic: model misuses external context: retrieves relevant docs but generates unsupported claims.

The 80–90% rule: most generation failures stem not from poor prompting, but from bad data. If your corpus lacks the answer, no amount of prompt engineering will fix it. Even AGI can’t recover from missing inventory, outdated docs, or conflicting values in the same document.

Tool use hallucinations are real and often ignored. When an agent:. - Invents a function name. - Uses incorrect parameters. - Calls a tool with invalid args.THIS is a hallucination. It’s not “prompt failure”; it’s a failure in context grounding. We now have open-weight models.

You can’t evaluate hallucinations in production using golden answers. You don’t have them. Instead, break evaluation into two independent questions:. - Was relevant information retrieved?. - Did the model use that information correctly?.Answering these doesn’t require ground.

How do tech companies and hr justify #100milliondollaroffers ? Or has every elite tech research read @AlexHormozi’s book 😝.

This is what I like about tech. A kid 18 years younger than you still calls you bro.

I can just do things without knowing it all 🔥.

🚀 The 3 Gulfs of LLM Pipeline Development (And how not to drown!) 🌊. Building LLM pipelines can feel a bit like trying to teach your GPS directions while blindfolded. You think you know where you're going, but somehow you end up in a Wendy's parking lot at 2 AM. You may.

Will MBA become masters of business (ai) agents?.

Those who crib about wrappers do not understand emergent properties.

So exciting; gonna try this out :).

I once backpacked SE Asia and Indian Subcontinent for 1.5 years. It was supposed to be 5 months but I fell.down the pragmatic dharma and Buddhism rabbit hole and went to several monasteries to meditate.