Zafda

Unhinged but self-aware

[Zafda @ZafdaOne]

@Dexerto Trillion dollar company with the tax planning skills of a college drug dealer.

[Xplained @oddedit]

Saw this line from @wealth_director today:
"Everyone you meet is smarter than you at something. Every conversation is a chance to learn something new. Stay curious."
It reminded me of an interview I took two years ago.

Two years back, I was interviewing a candidate for a global product-based firm. She had mostly worked on the operations side at a local bicycle manufacturer in Kenya. For the past two years, she had been trying to break into Product Management and learning through online videos, blogs, and product teardown articles.

The interview started smoothly. She came across as grounded and genuinely curious. But within the first 10 minutes, it became clear that she needed a bit more polish when it came to product thinking. For instance, she struggled to apply basic metrics like Average Order Value and conversion rate to real-world scenarios. She knew the definitions, but connecting the dots in a live product case was tough for her.

Most of our discussion revolved around how she would approach a product launch, diagnose a dip in conversions, or build something new for a user segment. Her fundamentals were okay, better on theory, less confident in live scenarios. But she was trying. You could see she had done the homework.

The company, on the other hand, had a high bar. They wanted someone who could own a product feature end-to-end, communicate crisply with leadership, and proactively flag blockers. They were expecting someone ready to hit the ground running.

As we wrapped up, I asked my usual final question, one I borrowed from the YC application:
"Tell me about a time when you hacked a non-computer system to your advantage."
She paused for a second, as if debating whether to say what was really on her mind. Then she smiled and said:
"There can be a huge difference in Uber fares if you just change the pickup location to the other side of the road." I chuckled, made a mental note, and we moved on.

Fast forward two years. I was in London, stepping out of a restaurant and trying to book a cab to my hotel. I suddenly remembered her hack. Out of curiosity, I checked the fare from my current location, then checked again from the other side of the road.
Oh my sweet lord.

The fare was 20% lower. Not just that, the ETA also dropped by 12 minutes because the other side of the road avoided a U-turn. Since then, every time I travel to a new city, I run this check. And more often than not, it works. Less fare, less time. Win-win.

As for the candidate, she joined the company, kept learning, and this year, she moved into a senior product role. The leadership is thrilled with her growth.

The point of the story? Everyone has something to teach you. It might not show up in the first five minutes or fit your expectations but if you stay curious, stay open, you will be surprised what sticks with you.

And sometimes, it will save you 20% on Uber.

[Xplained @oddedit]

Bias in AI

A team once built a model to help pick resumes. They gave it thousands of examples to learn from. At first, it seemed to work really well. But then someone noticed something strange. It was favouring certain kinds of names more than others. Nobody told it to do that. It just picked it up from the data.

That’s what bias in AI looks like.

The model wasn’t trying to be unfair. It just learned from patterns in the past. And if the past was already unfair, the model copies that without knowing any better.

AI doesn’t understand right or wrong. It doesn’t know what’s fair. It just sees what has happened before and assumes that’s how things should continue. So if one group got more opportunities in the past, the model keeps pushing those same results forward.

It’s like teaching a kid by only showing them old, one-sided stories. Of course they’ll grow up with a skewed view. And it’s not just about fairness. Bias makes the model worse. It performs badly for people it hasn’t seen much in the data, or it gets things completely wrong for certain groups. The worst part is, it often sounds confident while being totally off.

So what can we do about it? We check our data more carefully. We try to make sure it's more balanced. We test how the model behaves for different people. And most importantly, we keep asking hard questions. Who is this model helping? And who might it be leaving behind? As AI becomes part of how we hire people, give loans, provide medical advice, and make big decisions, we need to make sure it’s not just smart but also fair.

If you’re curious, try digging deep into “bias in AI” or “fairness in machine learning.” It’s easier to understand than you might think, and it’s one of the most important things we need to get right.

[Xplained @oddedit]

Hallucinations in AI

Yesterday I asked chatGPT to provide me the summary of a Japanese horror movie that I haven't watched, it provided a breakdown. The characters, the plot and even the conclusion. For a moment, I actually believed it but a few google searches and I realized that's its just making things up. Impressive Delivery. Zero truth.

This is Hallucination. Not in biology, but in AI. Its when the model confidently outputs wrong information. And yes, its not because of any wrong intention, its simply because the model has been trained not to choke at any point and it doesn't know the truth. Its almost like the college viva except the facts that the tables has turned and the examiner being you.

More than often, these answers gets past unnoticed. People make perceptions about things that are totally wrong. You ask “Where should I go in Wakanda?” and it responds “You should visit Golden City, the heart of the nation”. Sounds right, feels right. But still fictional. And the scary part is it offered me to provide with a planner for a 2 day trip to Wakanda.

But why does this happen? Because at times, it doesn't know the truth and has been modeled to predict the next word based on patterns. So, when the context is shaky, it matches patterns and go freestyle. And being freestyle is karaoke singers with bad memory - tuned in but off track.

Fixing this, we bring in Grounding. In simpler words, we ask AI to go search Wikipedia and then provide the answer. Retrieval Augmented Generation (RAG) is another upgrade, which pulls relevant documents before answering. Better prompts are also helpful when you ask a very specific question and not vague.

As AI gets louder and more and more people rely on the same for everyday information, we need to tune its lyrics. We are here not just to build smarter bots, build honest ones. Curious to dig deeper? Look up “AI grounding” or “RAG architecture.” It’s less intimidating than it sounds.

[Zafda @ZafdaOne]

Fallout (7.8/10) reinvents the wasteland with 65M viewers. TLOU (8.2/10) nails the game’s soul, 8 Emmys. Which game-to-series rules?

[Zafda @ZafdaOne]

@DiscussingFilm from island to volcano, ghost gonna roast this time as well.

[Zafda @ZafdaOne]

@SherryYanJiang Best of luck! +1 from us.

[Zafda @ZafdaOne]

@kimmaicutler Or you could buy X and then negotiating is not required.

[Zafda @ZafdaOne]

Philip. K Dick predicting Generative AI back in the 60s

[Zafda @ZafdaOne]

@DiscussingFilm This is gonna hurt so good.

[Zafda @ZafdaOne]

For 2,000 years, Europe’s had a stranglehold on the papacy. Out of 266 Popes:

1. Europe: 253
2. Asia: 9
3. Africa: 3
4. Americas: 1

The Vatican’s next chapter may finally turn the page.

[Zafda @ZafdaOne]

@LuisvonAhn @duolingo Duo preparing to threaten my family if I don't know what en passant is

[Zafda @ZafdaOne]

@Dexerto when you want the experience but not the commitment.

[Zafda @ZafdaOne]

@screenrant buffy is here to slayyyy

[Zafda @ZafdaOne]

@SherryYanJiang for content creators -> gold
for running ads -> vanity

[Zafda @ZafdaOne]

@DiscussingFilm I know -> I still know -> I'll always know -> I know

[Zafda @ZafdaOne]

@DiscussingFilm thanks for all the nightmares Jack. They were all so beautiful.

[Zafda @ZafdaOne]

@IGN give it up guys, this episode was outstanding. You can keep your 6/10.

[Zafda @ZafdaOne]

@ben_m_somers @garrytan Schools should always be important. You think it has ever been about what subjects they teach?

[Zafda @ZafdaOne]

@alexisohanian @TGL Dreams are fun. For everything else, it must be interactive.

[Zafda @ZafdaOne]

@alexisohanian We bake our own choregs and the best part is yet to come. Grandpa’s recipe optimising for health and strength. Go save the world with vibe coding and a bite of our coregs.