TravnikovDev

Hot take: Gmail isn’t secretly feeding Gemini your private emails. The scary posts made for great clicks - the boring truth didn’t.

My AI research agent pulled the receipts from Google’s Workspace privacy hub, plus Snopes and Malwarebytes’ corrections. Here’s the plain-English version.

• No, there was no automatic opt-in that let Google train generative AI on your inbox. For business/edu accounts, Google says your content isn’t used for gen-AI training without permission. Consumer Gmail got swept into a rumor in late 2025 that fact-checks knocked down. Nobody flipped your settings behind your back.
• Yes, Gmail reads message content to run long-standing features. That’s spam filtering, Smart Compose/Reply, category tabs, and those helpful summary cards. That processing powers the product - it is not the same as training a giant text generator on your private emails.
• Turning off Smart features stops those conveniences, not an AI training pipeline. You can still use Gmail just fine.
• The spell-check scare

TravnikovDev

You don’t need a Delaware fairy godmother to raise venture. In most emerging hubs, money moves on local rails - and it works.

I had my AI research agent pull how early rounds actually close across India, SEA, Africa, and LatAm. The pattern is repeatable: same economic logic as the US, but local paperwork, local regulators, local exit paths.

What this really looks like

• Early money uses straight equity, convertible notes, SAFE-like docs, and more recently revenue-based financing. Terms rhyme with Silicon Valley, but the wrappers are local. Typical priced seed or Series A sells about 10-30 percent. Exits are mostly acquisitions or secondaries. IPOs are a nice poster, not your plan.

Regional flavors

• India - priced rounds use CCPS; pre-seed often iSAFE that converts into CCPS. Foreign convertibles allowed with RBI filings. SEBI AIFs write many of the checks.
• Singapore/SEA - VIMA templates standardize term sheets and a CARE convertible. Ops often sit in Indonesia or Vietnam, but governance stays under Sing

TravnikovDev

Your post goes live and strangers like it in 60 seconds. No, you’re not special - the machine is testing you.

I dug in with my AI research agent and the receipts are clear 🕵️ - auto-likes and auto-comments exist, they’re banned, and they still slip through because some of it looks human and the detection isn’t perfect.

First, why those instant non-follower impressions happen at all - LinkedIn test-distributes every new post to a small circle beyond your followers. If people pause or comment, it widens the pipe fast. LinkedIn’s own engineers have written about this - dwell time matters: https://engineering.linkedin.com/blog/2020/improving-feed-relevance-with-dwell-time

How the fake gravy gets poured on top:

• Engagement pods - groups coordinate early likes/comments to juice reach. Think quick, clustered reactions, often with generic comments. Primer and pros-cons here: https://www.podawaa.com/blog/linkedin-engagement-pods and here: https://expandi.io/blog/linkedin-engagement-pods/
• Automation suites with A

TravnikovDev

If GitHub Gists could juice rankings, every casino and crypto site would be #1 by lunch. They’re not. And there’s a reason.

I had my AI research agent pull the raw signals on this, and the pattern is boring in the best way: gists get indexed, links get neutered, and spam gets ignored.

How it actually works:

• Gists are indexable. You can find them in Google. But they mostly show up for developer queries, code snippets, and very specific technical stuff.
• Links inside gists are almost always tagged nofollow/ugc. Translation: Google treats them as hints, not votes. You might get discovery. You won’t get trust.
• Gists rarely hold page-1 real estate for commercial keywords. For most niches, they’re background noise.

How spammers use it:

TravnikovDev

Stop waiting for your third startup. VC doesn’t reward scar count - it rewards speed, traction, and problems that burn cash to beat the clock.

2025 is not 2023. The market is pickier but alive. Global funding climbed back above $300B in 2024 and Q2/Q3 this year landed near $90-100B each, with AI drinking from the biggest hose. Small US rounds fell to roughly half of deals - the bar rose, not vanished.

I had my AI research agent pull NVCA, Carta, Crunchbase, Stanford - the numbers don’t back the “third-time” fairytale. Most companies start bootstrapped, and most never raise at all. Prior success helps a bit, sure, but there’s no magic unlock at startup three.

What VC is actually for: gasoline on a working fire when time-to-market, network effects, or capex outrun bootstrapping. Not life support.

Early-stage reality check:

• US seed is roughly $2-3M on about $15-16M pre-money.

TravnikovDev

A new type of profession is developing: AI tutors for models.

Here’s the twist - if you’ve ever nitpicked a code review or triaged a messy bug queue, you already qualify. You literally get paid to teach robots how to behave.

I had my AI research agent pull the raw data across Reddit worker threads and platform pages. The signal is noisy, but the pattern is clear: RLHF work is a real side hustle with real money for people who can write, reason, or code.

Top 10 platforms not on the usual list

• Outlier AI - expert RLHF across writing, math, code. High activity, best pay.
• DataAnnotation.tech - evals for writing, safety, coding. High activity, selective onboarding.
• Amazon MTurk - find AI eval, ranking, response rating. Always on, quality varies.

TravnikovDev

AI side hustles aren’t dead—they just got pickier. The bots still need babysitters, and they’ll pay you to judge them. 💸

My AI research agent pulled the receipts from papers and Reddit. The job is called RLHF: you rank model replies, write better ones, and flag safety issues. Those choices train a reward model; then the model gets nudged toward what humans prefer.

How the money works: expect an application, a tough exam, NDAs, and QA checks. Pay ranges from decent to “why am I doing this,” and task flow swings from feast to famine. Pro tip: don’t marry one vendor—keep 2–3 warm.

Quick hits from the trenches:

• TELUS: steadier search/eval hours if you pass a brutal exam; still cyclical.
• Surge AI: higher rates and real RLHF work; sporadic and selective.
• TaskUs: more like a job—stable schedules, metrics-heavy, location-dependent.

TravnikovDev

Good News No One Talks About (as usual)

I saw a viral claim that the Aral Sea “rose by 42%” and my skeptic mode kicked in. I asked my research bot to dig in Kazakh first, then Russian and English. The punchline: it’s not the water level that jumped 42%, it’s the water volume in the North Aral Sea - the Kazakh part - now around 27 billion cubic meters. Big deal anyway. Salinity dropped almost fourfold. Fish came back. The map looks less like a scar.

Here’s the part that surprised me: this “miracle” is mostly plumbing. The Kok-Aral Dam acts like a smart plug that keeps the Syr Darya’s water in the northern bowl long enough to build volume and dilute salt. Not flashy. Just gates, timing, and a river.

And it took logistics, not slogans. In 2024, Kazakhstan pushed a record 2.6 billion cubic meters into the North Aral, dredged channels, and tuned seasonal operations. Cross-border coordination mattered too. Some years bring more snowmelt, some less - you ride the flow you get.

Where this works: the North Aral. Y

TravnikovDev

Stop doomscrolling “Gemini killed ChatGPT” takes

I kept seeing posts claiming “Gemini jumped from 450M to 650M in a month” and “Gemini is beating ChatGPT.” Looked fishy. So I asked my bot-researcher to pull primary sources, not vibes.

Short answer: there’s no credible source for that one-month leap. Alphabet did say 650M+ MAU in Q3 2025. Earlier public numbers were ~350M in March 2025. That’s strong growth over months - not a magic +200M in 30 days. If a post can’t point to earnings materials, I bin it.

Who’s “winning”? Depends what you actually do. Arena-style rankings don’t show a universal champion. In practice: Gemini shines when I need absurd context or native video/audio in - long docs, meeting recordings, mixed media. GPT-4.1/4o feels steadier for coding reliability and snappy back-and-forth. I reach for each like different lenses.

Trade-off worth noticing: giant context is a freight train - powerful, but it takes longer to start and stop. GPT feels like a scooter - smaller trunk, faster through ci

TravnikovDev

5 numbers I can’t unsee

I got stuck on a weird question and asked my ChatGPT researcher to sanity-check it: which numbers are so upside down they change how I look at the world? The rabbit hole was worth it.

1. Sharks - fear is upside down 🦈
   ISAF confirms 4–10 unprovoked fatalities per year recently. Humans kill on the order of 70–150 million sharks annually. Culture says “Jaws.” Data says “us.” Caveat: the shark-kill range is wide because of illegal/unreported catches.

2. Chickens - ~200 million per day
   FAO headcounts imply 72–75 billion chickens slaughtered per year. Scale is the story. Thought experiment: if 8B people ate a 100 g chicken portion daily, we’d need roughly 400–550 million chickens per day depending on yield. Reality is closer to 200M because diets vary and a lot of food is wasted.