Weekly AI Updates (What's New Wednesday)·February 2026·5:49:49

Weekly AI Updates — February 2026 digest (4 episodes)

Karan Rana Host, Weekly AI Updates

The short version

Karan ran a two-part Google Flow (AI filmmaking) deep dive across Feb 4 and Feb 11, covering extend/jump clip-stitching, scene building capped near 2.5 minutes, and ingredients-to-video for locking up to 3 characters or props for consistency
AI browsers and agentic features accelerated all month: Gemini folded into Chrome, ChatGPT Atlas tested, and Grok, ChatGPT, and Claude all pushed agent/automation upgrades (Grok 4.2 'heavy' multi-agent mode, Claude Sonnet 4.6 autonomous workflows, Manus personal agents)
Anthropic dominated business news: a $30B raise ($380B valuation), Claude Opus 4.6 plugins rattling Indian IT stocks, reported Claude use in a Venezuela military strike, an AI safety researcher's resignation, and a new Goldman Sachs enterprise deal
Feb 18 was a dedicated, tool-free session on AI's environmental cost: hundreds of billions in data-center capex from Amazon, Alphabet, Meta, Microsoft, and Oracle plus Indian conglomerates, a triple water/carbon burden, and practical per-user efficiency tips
Feb 25 covered India's AI Impact Summit in New Delhi (Modi, Sam Altman, Sundar Pichai) and a live showdown between sovereign Indian model Sarvam AI (Bulbul text-to-speech, 22 Indic languages) and ElevenLabs
Smaller recurring threads: an AI-only social network called Moltbook going viral, a Meta patent for posthumous 'digital afterlife' AI social accounts that the class largely rejected, and Perplexity's multi-model 'council' verification approach
Tools sighted this month: Google Flow (VO 3.1, Nano Banana Pro); Gemini 3 / 3.1 Pro; Gemini in Chrome; ChatGPT Atlas; ChatGPT (Go, Plus, 5.3 Codex, Prism); Grok / Grok 4.2 / Grokipedia; Claude (Opus 4.6, Sonnet 4.6); NotebookLM; Perplexity; Manus; ElevenLabs; ByteDance desktop assistant; Sarvam AI (Bulbul, Vision, Translate, Samvad); Higgsfield AI; Zapier; Lyria 3; Moltbook

The concepts

AI Filmmaking with Google Flow (Extend, Jump, Ingredients)

Google Flow is Google's AI filmmaking tool (Gemini for reasoning, Nano Banana Pro for images, VO 3.1 for video) that Karan demoed across sessions on Feb 4 and Feb 11. Clips top out at 8 seconds each; 'extend' continues the last frame into the next shot for seamless motion while 'jump' ('cut to') starts a disjoint new shot, and both are stitched together in a Scene Builder capped around 2.5 minutes per scene. The 'ingredients-to-video' workflow lets you save up to three character or prop images as locked 'ingredients' so the same actors and props stay visually consistent across otherwise separate generations.

Extend continues the last frame of a clip into a new 8-second shot for seamless action, while Jump ('cut to') starts a fresh disjoint shot; each clip via VO 3.1 fast maxes out at 8 seconds (0:45:57, l3373865)

Scene Builder stitches clips into one scene capped at roughly 2.5 minutes, after which a new project must be started to continue the story (1:04:14, l3373865)

Ingredients-to-video locks up to 3 character or prop images (made with Nano Banana Pro) as reusable 'ingredients' so the same actors and props stay consistent across separate video generations (0:41:28, l3412215)

Live demo showed strong character consistency across shots but exposed limits: multi-character dialogue and lip-sync assignment are unreliable, and Google Flow currently recommends one dialogue per shot (1:00:12, l3412215)

Free tier gives 100 starting credits plus 50 new credits every morning that expire within 24 hours, a habit-forming daily-engagement mechanic (1:26:24, l3373865)

Environmental Impact of AI and Data Centers

On Feb 18 the class paused tool demos for a dedicated session on the environmental cost of AI, walking through public data-center capex commitments from Amazon, Alphabet, Meta, Microsoft, Oracle, and Indian conglomerates like Adani, Reliance, and Tata. The session traced a triple water/carbon burden -- ultrapure water to fabricate AI chips, chilled water or immersion cooling to run servers, and electricity often drawn from fossil-fuel ('dirty') grids -- and noted that current regulation (e.g., the EU AI Act) measures training energy but not the inference energy used by end users. It closed with practical, individual-level efficiency tips rather than policy prescriptions.

Big-tech data-center capex commitments cited: Amazon $200B, Alphabet $180B, Meta $125B, Microsoft $120B, Oracle $50B; Indian conglomerates Adani roughly $100B, Reliance and Tata roughly $50B each (0:40:43, l3412216)

Data centers carry a triple water/energy burden: ultrapure water to fabricate AI chips, chilled water or immersion cooling to run servers, and electricity from often fossil-fuel ('dirty') power plants (0:48:56, l3412216)

Net-zero pledges were described as increasingly unreachable, with AI data-center emissions projected toward roughly 40 million metric tons of carbon by 2030 (0:50:58, l3412216)

EU-style AI regulation currently measures training energy but not inference energy, so per-query usage by billions of end users goes largely unreported (1:01:03, l3412216)

Practical tips given to individuals: write one precise prompt instead of ten vague ones, default to light/mini models for simple queries, delay 4K/high-res exports until the final version, and toggle off unused AI copilots in everyday software (1:05:06, l3412216)

Sovereign Indic Voice AI: Sarvam vs ElevenLabs

Sarvam AI is a Bangalore-based, IIT-Bombay-adjacent Indian startup positioning itself as a 'sovereign' AI model whose data and training stay inside India rather than round-tripping to US servers. Its Bulbul text-to-speech model targets Indic languages (22 planned, 11 live at demo time) plus English, and was benchmarked live against ElevenLabs (70+ languages) on pronunciation, pricing, and a PDF/poster layout-preserving translation tool called Sarvam Vision.

Sarvam AI, incubated near IIT Bombay, positions itself as a 'sovereign' AI model -- data trained and processed inside India rather than round-tripping to US servers (0:32:18, l3412217)

Sarvam's Bulbul text-to-speech model targets 22 Indic languages (11 live at demo time) plus English, versus ElevenLabs' 70+ languages including Spanish and Chinese (0:36:21, l3412217)

Sarvam prices pay-as-you-go with non-expiring credits (about 10,000 INR for 11,000 credits) instead of ElevenLabs' subscription/rollover model (0:50:31, l3412217)

Sarvam Vision was demoed translating posters and PDFs into multiple Indian languages while preserving layout and formatting, not just extracting text (0:44:27, l3412217)

A live side-by-side Hindi/Kannada pronunciation test was mixed: the class rated Sarvam's output favorably, but a follow-up ElevenLabs generation of the same Hindi line was judged cleaner, with Sarvam mispronouncing the final word (1:02:44, l3412217)

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.

01AI Filmmaking with Google Flow (Extend, Jump, Ingredients)Google Flow is Google's AI filmmaking tool (Gemini for reasoning, Nano Banana Pro for images, VO 3.1 for vi…›

Scene Builder stitches clips into one scene capped at roughly 2.5 minutes, after which a new project must be started to continue the story (1:04:14, l3373865)

Free tier gives 100 starting credits plus 50 new credits every morning that expire within 24 hours, a habit-forming daily-engagement mechanic (1:26:24, l3373865)

02Environmental Impact of AI and Data CentersOn Feb 18 the class paused tool demos for a dedicated session on the environmental cost of AI, walking thro…›

Net-zero pledges were described as increasingly unreachable, with AI data-center emissions projected toward roughly 40 million metric tons of carbon by 2030 (0:50:58, l3412216)

EU-style AI regulation currently measures training energy but not inference energy, so per-query usage by billions of end users goes largely unreported (1:01:03, l3412216)

03Sovereign Indic Voice AI: Sarvam vs ElevenLabsSarvam AI is a Bangalore-based, IIT-Bombay-adjacent Indian startup positioning itself as a 'sovereign' AI m…›

Sarvam AI, incubated near IIT Bombay, positions itself as a 'sovereign' AI model -- data trained and processed inside India rather than round-tripping to US servers (0:32:18, l3412217)

Sarvam's Bulbul text-to-speech model targets 22 Indic languages (11 live at demo time) plus English, versus ElevenLabs' 70+ languages including Spanish and Chinese (0:36:21, l3412217)

Sarvam prices pay-as-you-go with non-expiring credits (about 10,000 INR for 11,000 credits) instead of ElevenLabs' subscription/rollover model (0:50:31, l3412217)

Sarvam Vision was demoed translating posters and PDFs into multiple Indian languages while preserving layout and formatting, not just extracting text (0:44:27, l3412217)

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 GPT / chart GBD	ChatGPT
mold book / board book / Moltbbook	Moltbook (the AI-agent-only social network)
Cortex	Codex (OpenAI's coding tool, referenced as 'ChatGPT 5.3 Codex')
Servam / Saram / Servant	Sarvam AI (Indian sovereign voice/LLM startup)
Higgs field angles version	Higgsfield AI (image/video camera-repositioning tool)
open clock creator, Peter	Uncertain -- likely refers to an open-source AI agent tool's creator reportedly hired by OpenAI; exact name/tool not clearly audible

True on recording day — verify before relying

T
h
i
s
d
i
g
e
s
t
r
e
f
l
e
c
t
s
A
I
n
e
w
s
a
n
d
p
r
o
d
u
c
t
c
l
a
i
m
s
a
s
n
a
r
r
a
t
e
d
l
i
v
e
i
n
f
o
u
r
W
e
e
k
l
y
A
I
U
p
d
a
t
e
s
s
e
s
s
i
o
n
s
d
u
r
i
n
g
F
e
b
r
u
a
r
y
2
0
2
6
(
F
e
b
4
-
2
5
)
.
C
o
m
p
a
n
y
v
a
l
u
a
t
i
o
n
s
(
e
.
g
.
,
A
n
t
h
r
o
p
i
c
a
t
$
3
8
0
B
,
t
h
e
S
p
a
c
e
X
/
x
A
I
c
o
m
b
i
n
e
d
e
n
t
i
t
y
a
t
$
1
.
2
5
T
)
,
p
r
i
c
i
n
g
,
c
r
e
d
i
t
a
l
l
o
t
m
e
n
t
s
,
m
o
d
e
l
v
e
r
s
i
o
n
n
u
m
b
e
r
s
(
G
e
m
i
n
i
3
.
1
P
r
o
,
C
l
a
u
d
e
S
o
n
n
e
t
/
O
p
u
s
4
.
6
,
G
r
o
k
4
.
2
,
C
h
a
t
G
P
T
5
.
3
C
o
d
e
x
)
,
a
n
d
f
e
a
t
u
r
e
a
v
a
i
l
a
b
i
l
i
t
y
(
r
e
g
i
o
n
a
l
r
o
l
l
o
u
t
s
,
f
r
e
e
-
t
i
e
r
l
i
m
i
t
s
)
c
h
a
n
g
e
q
u
i
c
k
l
y
a
n
d
s
h
o
u
l
d
b
e
i
n
d
e
p
e
n
d
e
n
t
l
y
v
e
r
i
f
i
e
d
b
e
f
o
r
e
b
e
i
n
g
t
r
e
a
t
e
d
a
s
c
u
r
r
e
n
t
f
a
c
t
.
S
o
m
e
n
a
m
e
s
a
r
e
h
o
s
t
p
r
o
n
u
n
c
i
a
t
i
o
n
s
o
r
A
S
R
t
r
a
n
s
c
r
i
p
t
i
o
n
s
o
f
f
a
s
t
-
m
o
v
i
n
g
n
e
w
s
i
t
e
m
s
(
e
.
g
.
,
"
M
o
l
t
b
o
o
k
"
,
t
h
e
O
p
e
n
A
I
h
i
r
e
r
e
f
e
r
r
e
d
t
o
a
s
a
n
"
o
p
e
n
c
l
o
c
k
c
r
e
a
t
o
r
"
)
a
n
d
m
a
y
b
e
i
m
p
r
e
c
i
s
e
-
-
c
r
o
s
s
-
c
h
e
c
k
a
g
a
i
n
s
t
p
r
i
m
a
r
y
s
o
u
r
c
e
s
.