DAILY NEWS

Stay Ahead, Stay Informed – Every Day

Advertisement
TECH & AI
AI Video Scripts That Actually Convert: The Workflow Top Creators Use
jackminion 0



Your AI is writing scripts that rank nowhere because it’s optimizing for grammatical perfection instead of viewer retention. Here’s how top creators are actually using models to generate 10x more content without sounding robotic.

I’ve watched creators spend $500/month on AI tools and still post videos that flatline at 200 views. The problem isn’t the tools. It’s how they’re prompting them.

Most creators copy-paste a generic prompt like “write a YouTube script about (topic)” and wonder why the output sounds like a corporate press release. Meanwhile, a small group of creators is generating 15–20 scripts per week, maintaining 65%+ average view duration, and doing it with a workflow most people aren’t discussing publicly.

The Script Generation Trap: Why AI-Written Videos Feel Soulless

YouTube data from creator briefings and retention analytics from channels like Think Media and MKBHD shows that average view duration drops roughly 23% when the first 30 seconds feel scripted and impersonal.

That number should concern you.

The trap works like this: you ask an AI to write a script, it produces something grammatically flawless, topically accurate, and completely devoid of personality. It starts with “In this video, we’re going to explore…” and your viewer’s thumb is already moving.

The problem is structural. Most AI models optimize for coherence and completeness—great for research reports, terrible for video hooks.

Here’s what happened: Marcus, who runs a 180K subscriber personal finance channel, switched to pure AI scripts last year. His click-through rate held at 8.2%, but average view duration collapsed from 54% to 31% in six weeks. Viewers clicked, heard the robotic cadence in the first 10 seconds, and left. CTR is vanity. Retention is revenue.

The fix isn’t abandoning AI. It’s training the model on how you think, not just what you want to say.

The Personality Injection Framework: Embedding Your Voice Without Manual Rewrites

Top creators scaling production aren’t using different tools. They’re using a three-layer prompting system called Voice DNA.

Layer 1: The Sample Bank

Before generating any script, feed the model 5–7 transcript excerpts from your best-performing videos—specifically moments that got the most comments or timestamps in comments. You’re not asking the AI to copy these. You’re establishing a behavioral reference.

Here’s the prompt structure:

“The following are excerpts from my highest-retention videos. Analyze the sentence length patterns, the way I transition between ideas, the specific types of analogies I use, and my cadence when making key points. (PASTE TRANSCRIPTS). Now write a script about (TOPIC) that maintains these structural patterns while covering these beats: (BEATS).”

The difference is immediate. Priya, who runs a 90K subscriber productivity tools channel, tested identical topics with and without the sample bank layer. Scripts with her Voice DNA prompt averaged 61% view duration. Without it: 38%.

Layer 2: The Contrarian Flag

Your best videos almost certainly contain a moment where you said something surprising. The comment section lights up. These moments are algorithmically valuable because they trigger emotional engagement.

Add this to every prompt: “Include one counterintuitive claim in the first 90 seconds that challenges conventional wisdom about (TOPIC). This should feel like something I personally discovered, not a generic hot take.”

Being mildly controversial is far less damaging than being forgettable.

Layer 3: The Unfinished Sentence Technique

Instruct the AI to leave 3–4 moments marked as (YOUR RIFF HERE). These are 10-second gaps where you insert something spontaneous when recording.

This breaks the robotic cadence pattern that both algorithms and human ears detect, while preserving the authentic improvisation that longtime viewers expect. The script becomes scaffolding, not a cage.

Platform-Specific Optimization: Structure That Actually Performs

A script that works on YouTube long-form will kill your TikTok account.

YouTube Long-Form (8–20 minutes)

The structure that’s performing now is the Problem-Proof-System loop. Open with a specific problem (“I was spending 4 hours per week editing” beats “editing is time-consuming”). Provide proof you’ve solved it (a number, result, or screenshot). Deliver the system in 90-second chunks.

Key prompt addition: “Structure each section to end with a forward-looking sentence implying the next section is necessary.” This creates the binge effect.

Channels doing this well—Matteo French on productivity, Ali Abdaal’s team on study techniques—see 55–70% average view duration on videos over 12 minutes.

YouTube Shorts and TikTok (Under 60 seconds)

These reward pattern interrupts, not information delivery: Unexpected hook (0–3s) → Bold claim (3–8s) → Fastest proof (8–25s) → Restatement with twist (25–45s) → CTA that feels like a new story (45–60s).

For short-form, add: “The script should feel unfinished, like I’m sharing a discovery I just made, not teaching a lesson I’ve known for years.”

Instagram Reels (The Hybrid)

Reels viewers tolerate more structure than TikTok but have less patience than YouTube Shorts. Use: 3-second visual hook, one bold sentence that works on mute (60% of Reels are watched muted initially), and a comment-bait ending.

Prompt your AI: “End with a question that has no obvious right answer but every viewer has an opinion on.”

Real Workflow: The Tool Stack Top Creators Use

Step 1: Claude for Structure

Claude (3.5 or later) handles structural architecture better than other models because it maintains longer contextual coherence. When building 15-minute scripts with multiple segments, it doesn’t lose the thread.

Use Claude for macro-level outline, transition logic, and argument structure. Feed it your Voice DNA samples first. Expect 2–3 iterations on structure.

Step 2: Custom GPT or Fine-Tuned Model for Tone

This is where personality lives and where most creators skip. Build a custom GPT trained on your transcript library. It handles sentence-level tone adjustments—how you phrase things, your filler-word patterns, your joke structure.

Workflow: get structure from Claude, paste into your custom GPT with: “Rewrite this in my voice, maintaining all structural beats.” This two-model workflow adds 8 minutes and produces noticeably more authentic output.

Step 3: Multimodal Validation

Run the script through text-to-speech (ElevenLabs with voice clone or native TikTok reader) before recording. Listen for moments that sound wrong—these are your (YOUR RIFF HERE) markers.

One 400K tech channel creator said this step cut re-record rate from 40% to under 10%.

Time breakdown: Full workflow—Voice DNA prompt, Claude structure, custom GPT tone pass, validation—takes 35–45 minutes per script. Manual scripting typically takes 2–3 hours. You’re faster with a higher quality floor.

Avoiding the Detection Penalty: What YouTube Actually Cares About

YouTube isn’t uniformly penalizing AI-generated content. They’re penalizing low-effort, undifferentiated content—and AI scripts are the fastest way to produce that at scale.

YouTube tracks “satisfaction signals”: comment sentiment, share rate, full vs. partial view ratio, and subscription conversions after watching. AI-generated scripts tend to produce low share rates and near-zero subscribe conversions because they don’t create the parasocial connection that drives growth.

The signal YouTube reads: if 1,000 people watch and none subscribe, the algorithm infers that while the title/thumbnail worked, the content didn’t create lasting interest. Distribution gets throttled accordingly.

Your AI scripts need moments that feel unreplicable—specific personal details, real numbers, opinions clearly yours. These aren’t decoration. They’re conversion events.

Add this to every prompt: “At three points, flag a (PERSONAL INSERT) marker where I should add a specific detail, anecdote, or real data point from my own experience. Place these at moments of highest emotional or logical weight.”

This also addresses detection tool concerns. Services like Originality.ai are used by brand partners and MCNs. Make your content genuinely personal, not just structurally varied.

Winning creators treat the model as a research assistant and structural partner, not a ghostwriter. They’re in the video. Their opinions are in the video. The AI helped them get there faster.

Action This Week

Don’t overhaul your entire workflow. Do this instead:

Take your three best-performing videos (by average view duration), run them through transcription (Descript or Otter.ai), and save transcripts in one document. That’s your Voice DNA sample bank.

Next time you generate a script, start with: “Here are transcripts from my three highest-retention videos. Before writing anything, identify five specific patterns: sentence length, transition style, analogy type, question placement, and emotional cadence. Then use those patterns for a script about (TOPIC).”

Run that prompt once. Compare the output to what you’ve been getting.

The money on the table isn’t views. It’s the compounding effect of a 20-percentage-point improvement in average view duration, applied across 52 weeks of content. That’s the gap between a channel that stagnates at 50K subscribers and one that hits 500K.

Your voice is the asset. AI is the production infrastructure. Understanding that distinction separates creators building something lasting.

Follow for more practical AI and productivity content.



Source link

TECH & AI
Anatomy of Duck DB for Python Developers
jackminion 0



Introduction – SQL without a Server

Pandas is widely used for data analysis and almost every data analyst or even data engineers utilize it for faster analysis with table like data structure called DataFrames.The drawback is that it suffers once the data goes beyond few GB’s and spinning up a Postgres or a Redshift is an overkill for quick analysis.Duck DB fills this gap with Zero-setup columnar SQL.

Getting Started – zero config, instant power

DuckDb is an open source OLAP database management system designed for analytics and for running within the same process as the application.It is lightweight, can work directly with data files in csv, parquet etc without needing a server.

Installation and first query

pip install duckdb – No ports to open, No configuration and No daemon

In-Memory and Persistent Database – Two Operating Modes

In-MemoryWhen DuckDB connection is created without specifying a file, a database lives entirely in RAM.

import duckdb
con = duckdb.connect() # or duckdb.connect(‘:memory:’)

Enter fullscreen mode

Exit fullscreen mode

All data is stored in RAM and no files are written to disk
Extremely fast reads/writes since there is zero I/O overhead.
Data is completely lost when connection closes.
No file locking or concurrency concerns

Persistent ModeWhen the user provides a location DuckDB can write the results to disk in .duckDb format.

con = duckdb.connect(‘my_database.duckdb’)

Enter fullscreen mode

Exit fullscreen mode

Tables,Schemas and indexes are persisted.
Uses a columnar storage format with compression and buffered I/O
Only one write connection at a time but multiple read connection are allowed.
Supports WAL(Write Ahead Logging) for crash recovery

Powerful Pattern

DuckDb allows you to mix both modes where user can start with in-memory and attach a persistent database or use copy/export to snapshot in-memory result to disk.

con = duckdb.connect()

#Query a CSV, transform it, save the result to a persistent file
con.execute(“””
COPY(SELECT region, SUM(sales) AS total FROM read_csv(‘data.csv’)
GROUP BY region
)
TO ‘results.parquet’ (FORMAT PARQUET)
“””)

Enter fullscreen mode

Exit fullscreen mode

Users gets the speed of In-Memory processing which accelerates the pipeline processing with an option to persist.

Reading files directly –CSV,PARQUET,JSON,Arrow,

Query CSV without loading into memory

Select * from read_csv(‘data_csv’, auto_detect=true);

Enter fullscreen mode

Exit fullscreen mode

-Auto detects delimiter, compression and data types-Handles malformed rows gracefully-Can read multiple CSVs at once read_csv(‘data/*.csv’)

Parquet

Select * from read_parquet(‘data.parquet’);
–even from S3 directly
Select * from read_parquet(‘s3://bucket/data/*.parquet’);

Enter fullscreen mode

Exit fullscreen mode

Exploits column pruning as it only reads columns you need
Leverages row group skipping using Parquet’s build in min/max stats
Native support for nested types(structs,list,maps)

JSON/NDJSON

SELECT * FROM read_json(‘events.ndjson’, auto_detect=true);

Enter fullscreen mode

Exit fullscreen mode

-AUTO INFERS schema from data-NDJSON(Newline delimited) streams efficiently line by line-Can unnest deeply nested JSON fields using DuckDB’s json_extract, UNNEST, or -> operators

Apache Arrow

import pyarrow as pa
arrow_table = pa.Table.from_pandas(df)
duckdb.query(“”SELECT * from arrow_table”””)

Enter fullscreen mode

Exit fullscreen mode

-Zero copy integration: DuckDB reads from Arrow memory without serialization-Ideal for pipelines where data never needs to touch disk

SQL Beyond Select

DuckDB is not just a query engine, it supports rich SQL that covers data transformation, creation, and some genuinely unique syntax extensions to available in most databases.

Full Suite of WINDOW Functions

Select
customer,
ordered_at,
amount,

— Running total
SUM(amount) OVER (PARTITION BY customer ORDER BY ordered_at) AS running_tot,

— Lag/lead comparisons
LAG(amount) OVER (PARTITION BY customer ORDER BY ordered_at) AS prev_amt,

— Percentile rank
PERCENT_RANK() OVER (ORDER BY amount) AS pct_rank,

— Named window reuse
FIRST_VALUE(amount) OVER w AS first_order
FROM orders
WINDOW w AS (PARTITION BY customer ORDER BY ordered_at);

Enter fullscreen mode

Exit fullscreen mode

DuckDB also allows the use of qualify clause which filters on window result without a subquery.

Select * From orders
QUALIFY ROW_NUMBER() OVER (PARTITION BY customer ORDER BY amount DESC) = 1;

Enter fullscreen mode

Exit fullscreen mode

PIVOT and UNPIVOT

Most databases make you write case when manually for PIVOTS.DuckDB does it natively.

–PIVOT- rows to columns
PIVOT orders on region USING SUM(amount) GROUP BY year;

–UNPIVOT- Column to rows
UNPIVOT sales_wide ON(q1,q2,q3,q4) INTO NAME quarter VALUE revenue;

Enter fullscreen mode

Exit fullscreen mode

MULTI DATABASE SQL

–Attach another DuckDB file
ATTACH ‘archive.duckdb’ AS archive;

— Cross-database join
SELECT a.*, b.region
FROM main.orders a
JOIN archive.customers b ON a.customer_id = b.id;

–Attach another database
ATTACH ‘postgres://user:pass@host/db’ AS pg (TYPE POSTGRES);
SELECT * FROM pg.public.users LIMIT 10;

Enter fullscreen mode

Exit fullscreen mode

DUCKDB+Pandas+Polars –Choosing your stack

DuckDB does not replace pandas or Polars it solves a problem which is niche.The sweet spot of the industry is to use DuckDB for SQL-shaped operations and pandas/polars for row level python logic.

Where Duck DB shines

Feature Engineering for ML: Window functions or group by’s for feature computation are often faster and more readable in DuckDB then pandas before handing it over to Sklearn or pytorch
Unit testing DBT models locally:DuckDB lets you run complete dbt project locally without a cloud warehouse providing fast feedback loop for data engineers.
Light weight ETL Pipelines: One can read raw parquet from S3, transform with SQL, write cleaned output back without any spark cluster or airflow jobs.

Conclusion

DuckDB lets you think in SQL for analytical tasks without worrying about infrastructure setup. Anyone using python can utilize duckdb for analysis of larger files where regular pandas will give headache.Given the advantages, it is important to know whare DuckDB should not be used which in case of concurrent writes,OLTP workloads and long running multi user services.

Reference-https://duckdb.org/docs/current/data/overview



Source link

TECH & AI
Properties of scroll-timeline: creating animations on scroll without JavaScript
jackminion 0



Stop Using JS for Scroll Animations: Meet Scroll-Timeline

Grab a coffee, friend. We need to talk about that heavy JavaScript library you are probably using just to make a header shrink or a progress bar move. It is 2026, and the days of hijacking the main thread with scroll listeners are officially over. We finally have scroll-timeline, and it is a total game-changer for both performance and developer sanity. Imagine creating complex parallax effects with the same ease as a simple hover transition.

How we suffered before

Remember the struggle? To create a simple parallax effect or a reading indicator, we had to attach an event listener to the window scroll. Then came the “scroll-jank” – that stuttering mess when the browser could not keep up with the JavaScript calculations and the rendering at the same time. We tried to fix it with requestAnimationFrame, debouncing functions, or bringing in heavy-duty libraries like ScrollMagic or GSAP. While those tools are powerful, they are often overkill for simple UI polish. We even spent time styling the scrollbar in all modern browsers just to make things look cohesive, but the logic remained bulky and JS-dependent. It was a lot of code for something that should have been native.

The modern way in 2026

Now, we have CSS Scroll-driven Animations. The core idea is simple: instead of an animation progressing over time (seconds), it progresses over scroll distance (pixels or percentage). Using scroll-timeline, we can define a named timeline on a scrollable container. Then, we link any element’s animation to that timeline using animation-timeline. It is declarative, it is readable, and most importantly, it runs off the main thread. If you have already mastered managing scroll behavior with overscroll-behavior, this is the natural next step in your CSS journey. You are no longer calculating offsets; you are just describing how things should look at the start and end of the scroll.

Ready-to-use code snippet

Here is a classic example: a reading progress bar that grows as you scroll down the page. Notice how we do not need a single line of script to make this happen.

/* 1. Define the animation as you normally would */
@keyframes grow-progress {
from { transform: scaleX(0); }
to { transform: scaleX(1); }
}

/* 2. Setup the scroll container and name the timeline */
body {
scroll-timeline-name: –reading-timeline;
scroll-timeline-axis: block; /* ‘block’ refers to the vertical scroll axis */
}

/* 3. Link the progress bar element to the scroll timeline */
.progress-bar {
position: fixed;
top: 0;
left: 0;
width: 100%;
height: 8px;
background: #ff4757;
transform-origin: 0 50%;
z-index: 100;

/* The magic happens here: no duration in seconds, but ‘auto’ */
animation: grow-progress auto linear;
animation-timeline: –reading-timeline;
}

Common beginner mistake

The most common pitfall is forgetting the animation-duration. Even though the animation is driven by scrolling and not time, the CSS specification still requires a duration value (set to auto or any time value like 1s) for the animation to actually initialize. If you omit it, your animation might just sit there doing nothing, leaving you scratching your head. Also, ensure your scroll-timeline-name is defined on an actual scrollable parent; if the container does not have overflow: auto or scroll (or it is the body), the timeline will not have any range to work with and your animation will stay stuck at the first frame.

🔥 We publish more advanced CSS tricks, ready-to-use snippets, and tutorials in our Telegram channel. Subscribe so you don’t miss out!



Source link