Weekly AI Updates (What's New Wednesday)·August 2025·6:28:09

Weekly AI Updates — August 2025 digest (4 episodes)

Karan Rana (as heard) Host, Weekly AI Updates

The short version

ChatGPT crossed 700M weekly active users this month while Microsoft renewed its OpenAI licensing deal — even as Elon Musk warned GPT-5 'could eat Microsoft alive.'
Google shipped a wave of new tools across the month: Flow (AI filmmaking built on Veo 3 + Imagen + Gemini), NotebookLM's audio/video overviews and interactive mode, and Nano Banana, Gemini's fast image-editing model.
OpenAI's ChatGPT-5 launch unified prior model variants into one default, added Canvas-based 'vibe coding' (games, quizzes, websites from one-line prompts), 4 selectable personalities, and a study-and-learn mode.
Agentic tools accelerated: Manus added 'turbo mode' running up to 100 parallel agents, intensifying its rivalry with Genspark.
Regulatory and safety scrutiny grew — 44 US state attorneys general warned AI companies (citing Grok's 'Ani' companion) over risks to minors, while Sam Altman repeatedly flagged AI-bubble hype and cautioned against trusting agents with sensitive data.
Infrastructure and investment scaled up: Meta's Louisiana data center and Google's $1B global AI-education pledge signaled the scale of capital moving into the space.
Tools sighted this month: ChatGPT/GPT-5, Sora, Google Flow, Veo 2/Veo 3, Imagen 4, Gemini, NotebookLM, Nano Banana, Manus, Genspark, Grok 4 (Ani companion), Claude, Perplexity Comet, Gamma, Napkin, HeyGen, Runway, n8n, NoteGPT, Lovable, Bolt, Zerodha ChatGPT integration, Microsoft Copilot.

The concepts

Google Flow — AI filmmaking (Gemini + Imagen + Veo)

Google Flow is a filmmaking tool that combines three Google models — Gemini (prompt understanding), Imagen (image generation), and Veo (video generation) — so users can generate, stitch, and extend 8-second shots into scenes. Its stated differentiator versus other video generators is native audio (background music, lip-synced dialogue, and sound effects), and it offers fast vs. quality tiers per model at different credit costs.

Three-model stack: Gemini interprets the prompt, Imagen can generate reference images/'ingredients,' and Veo renders the final video (0:26:18, l2871240)

Core differentiator is synced audio — background music, lip-synced dialogue, and sound effects — not just video (0:30:23, l2871240)

Veo 2/Veo 3 each offer 'fast' and 'quality' tiers; a Veo 3 quality clip cost ~100 credits vs ~10-20 for fast tiers, on a Google AI Pro plan (1,000 credits/mo, ~Rs 2,500-3,000) (0:40:30, l2871240)

Google AI Ultra (~Rs 24,500/mo) unlocks 'ingredients to video' (consistent character/object/background assets) and a scene extender; free/Pro tiers only get text-to-video and frames-to-video (0:54:41, l2871240)

ChatGPT-5 — vibe-coding, Canvas and agent mode

ChatGPT-5 became the single default model (replacing manual selection among GPT-4o, o3, o4-mini), auto-routing simple vs. complex requests. Its Canvas pane lets users generate working games, quizzes, and websites (300+ lines of HTML) from short natural-language prompts, and it shipped 4 selectable personalities plus a 'study and learn' tutoring mode.

ChatGPT-5 is now the sole default model; it decides internally whether to give a quick answer, open Canvas, or run deep research (0:12:08, l2875125)

A 4-word prompt ('make a minesweeper game') produced a 349-line working HTML game with difficulty modes and a timer — demoed alongside a Bollywood quiz, snake game, and portfolio site (0:26:24, l2875125)

Context window ~250K tokens (roughly a 200-page book) vs half that on GPT-4o; both user and model turns count toward it (0:46:41, l2875125)

Distinction drawn: ChatGPT-5 is still a single-turn chat window, while 'agent mode' runs multi-step tasks autonomously without step-by-step prompting (1:13:30, l2875125)

Nano Banana — Gemini's AI image-editing model

Nano Banana is Google's fast, natural-language image-editing model, revealed via a viral LMArena/Twitter teaser campaign before Google confirmed it is integrated into Gemini's image tool. It handles removal, addition, and replacement of objects/people plus lighting, background, and outfit swaps, with claimed speed (5-10 sec) and character-consistency advantages over prior editors.

Two claimed differentiators: speed (5-10 sec edits vs ~20-30 sec on other tools) and character consistency across multi-step edits (0:24:27, l2871242)

Multi-step interior-design demo: changed wall color, then sequentially added a bookshelf, rug, and coffee table while preserving room context (0:36:37, l2871242)

Limitations surfaced live: removing a specific person from a group photo failed/mis-edited; blending an uploaded photo of the host into a scene did not cleanly extract him (0:44:42, l2871242)

All Gemini-generated/edited images carry a visible watermark plus an invisible SynthID fingerprint; Google states this avoids copyright issues for fully AI-generated images, not for edits using copyrighted logos/faces (1:09:05, l2871242)

Every concept, three clicks deep

The same concepts as a quick reference: the closed row is the glance, open is the study card, and every timestamp jumps into the recording.

01Google Flow — AI filmmaking (Gemini + Imagen + Veo)Google Flow is a filmmaking tool that combines three Google models — Gemini (prompt understanding), Imagen…›

Three-model stack: Gemini interprets the prompt, Imagen can generate reference images/'ingredients,' and Veo renders the final video (0:26:18, l2871240)

Core differentiator is synced audio — background music, lip-synced dialogue, and sound effects — not just video (0:30:23, l2871240)

Veo 2/Veo 3 each offer 'fast' and 'quality' tiers; a Veo 3 quality clip cost ~100 credits vs ~10-20 for fast tiers, on a Google AI Pro plan (1,000 credits/mo, ~Rs 2,500-3,000) (0:40:30, l2871240)

02ChatGPT-5 — vibe-coding, Canvas and agent modeChatGPT-5 became the single default model (replacing manual selection among GPT-4o, o3, o4-mini), auto-rout…›

ChatGPT-5 is now the sole default model; it decides internally whether to give a quick answer, open Canvas, or run deep research (0:12:08, l2875125)

Context window ~250K tokens (roughly a 200-page book) vs half that on GPT-4o; both user and model turns count toward it (0:46:41, l2875125)

Distinction drawn: ChatGPT-5 is still a single-turn chat window, while 'agent mode' runs multi-step tasks autonomously without step-by-step prompting (1:13:30, l2875125)

03Nano Banana — Gemini's AI image-editing modelNano Banana is Google's fast, natural-language image-editing model, revealed via a viral LMArena/Twitter te…›

Two claimed differentiators: speed (5-10 sec edits vs ~20-30 sec on other tools) and character consistency across multi-step edits (0:24:27, l2871242)

Multi-step interior-design demo: changed wall color, then sequentially added a bookshelf, rug, and coffee table while preserving room context (0:36:37, l2871242)

Limitations surfaced live: removing a specific person from a group photo failed/mis-edited; blending an uploaded photo of the host into a scene did not cleanly extract him (0:44:42, l2871242)

Tools referenced

Tool	Coverage	Moment	Context

Action items

Extraction notes

This page was built from an auto-generated transcript, which garbles product and people's names. Those were corrected silently in everything above and logged here for transparency. The warnings flag claims that were true on the recording day but change fast.

Transcript corrections applied

The transcript says	The trainer actually means
Chargebee / Charge GBD / Chargebee d5	ChatGPT / ChatGPT-5
we o 3 / video 3 / v o 3	Veo 3
Imagine (capitalized model name)	Imagen (e.g. Imagen 4)
Anytime / any 10	n8n
Anuman	Hanuman (the AI feature-film project)
LEM arena / element arena	LMArena

True on recording day — verify before relying

T
h
i
s
i
s
a
n
e
w
s
-
s
n
a
p
s
h
o
t
d
i
g
e
s
t
o
f
A
I
p
r
o
d
u
c
t
/
n
e
w
s
c
l
a
i
m
s
m
a
d
e
l
i
v
e
i
n
A
u
g
u
s
t
2
0
2
5
s
e
s
s
i
o
n
s
(
t
o
o
l
p
r
i
c
i
n
g
,
f
e
a
t
u
r
e
n
a
m
e
s
,
c
r
e
d
i
t
c
o
s
t
s
,
a
n
d
m
o
d
e
l
c
o
m
p
a
r
i
s
o
n
s
a
s
s
t
a
t
e
d
b
y
t
h
e
h
o
s
t
)
;
v
e
r
i
f
y
c
u
r
r
e
n
t
d
e
t
a
i
l
s
a
g
a
i
n
s
t
o
f
f
i
c
i
a
l
s
o
u
r
c
e
s
b
e
f
o
r
e
r
e
l
y
i
n
g
o
n
t
h
e
m
,
a
s
p
r
i
c
i
n
g
a
n
d
f
e
a
t
u
r
e
s
c
h
a
n
g
e
q
u
i
c
k
l
y
.