Weekly AI Updates (What's New Wednesday)·December 2025·9:05:27

Weekly AI Updates — December 2025 digest (5 episodes)

Karan Rana Host, Weekly AI Updates · Raghuveer ("Arvi") Guest host, Weekly AI Updates (Dec 31 episode; independent AI generalist/consultant)

▶local recording · LMS lesson

The short version

Notion AI matured into an agentic assistant this month, building databases, Gantt charts, and WhatsApp summaries from a single prompt and pulling web research into existing pages (0:56:25, l3131663).
Hedra positioned itself as an all-in-one creative studio combining avatar image generation, 11Labs-powered voice, and video/lip-sync, including a novel doodle-to-image/video feature (0:42:52, l3151699).
AI music generation went mainstream via Suno and 11 Labs Music, both producing full songs from text prompts in seconds, but both strictly gate commercial/monetized use behind paid tiers and ban celebrity-name or celebrity-voice prompts (1:20:16, l3198451; 1:34:15, l3217870).
Infrastructure and geopolitics kept surfacing: Microsoft committed $17.5B to India and sovereign AI, the Pentagon began preparing for AGI as a near-term risk, and NVIDIA trained a small LLM aboard an orbiting satellite (0:29:42, l3198451; 0:31:44, l3198451).
Everyday creative tools got AI upgrades: ChatGPT added Photoshop-style editing and GPT-5.2, Gemini added image markup/annotation and live speech translation, and YouTube shipped a Create app to compete with CapCut (0:33:37, l3217870).
The Dec 31 year-end recap traced 2025 from DeepSeek R1 (Jan) to Gemini 3 (Dec) and previewed 2026 themes: NotebookLM-as-a-Gemini-source, agentic tools like ClickUp Super Agents and MiniMax 2.1, and OpenAI's self-strengthening Atlas prompt-injection defense (0:12:58, l3237330; 0:33:14, l3237330).
Tools sighted this month: Notion AI; Claude Opus 4.5; Claude Interviewer; Grok; LTX Studio (Retake); Flux 2; Loart; Perplexity (virtual try-on, Nano Banana Pro, Sora 2 Pro, Kimi K2 Thinking); ChatGPT/GPT-5.2/GPT-image-1.5; Gemini 3 (Deep Think, Flash, image markup); Runway Gen 4.5; Hedra; Suno; 11 Labs Music; YouTube Create; Meta Sam Audio; Meta Vibes; Google Translate; Spatial; NotebookLM; ClickUp Super Agents; OpenAI Atlas; MiniMax 2.1; Alibaba Tongyi FunAudioChat; Krea; BeatOven.ai; Gamma; Sunday Robotics Act-1/MEMO.

The concepts

All-in-one AI creative studios

Instead of juggling separate tabs for image, voice, and video generation, tools like Hedra bundle character/avatar creation, 11Labs-powered voice, and video/lip-sync into one workflow, and Notion AI similarly bundles databases, calendars, and web research behind one agent prompt. The trend is toward fewer, more consolidated multimodal platforms rather than single-purpose tools.

Hedra bundles image generation, 11Labs-powered voice, and video/avatar creation into one "studio" workflow instead of three separate tools (0:42:52, l3151699)

Users can doodle a rough sketch and convert it into a styled image or video via Hedra's sketch-to-image feature (0:38:48, l3151699)

Hedra supports character consistency by saving a generated avatar as a reusable character ID reused across clips, up to roughly two stitched clips (1:01:06, l3151699)

Notion's AI agent can create, populate, and edit databases, Gantt charts, and calendars from one prompt, drawing on GPT, Claude Opus, and Gemini 3 models on the backend (0:56:25, l3131663)

Notion AI can be asked to "ruthlessly critique" existing pages to surface content or workflow gaps rather than just generate new content (1:06:32, l3131663)

AI music generation and commercial-use gray areas

Suno and 11 Labs Music both turn a short text prompt into a complete song (auto-written lyrics plus instrumentation) in under a minute, and both offer persona features for voice/style consistency across tracks. Neither allows true voice cloning of a specific person or frame-level in-app editing, and commercial monetization is explicitly tied to paid subscription tiers rather than to who technically generated or owns the output.

Suno and 11 Labs Music generate full tracks, including auto-written lyrics and instrumentation, from a single text prompt in roughly 5-10 seconds (0:53:59, l3198451; 0:53:48, l3217870)

Commercial or monetized use of AI-generated music requires a paid plan; free-tier tracks cannot legally be monetized even for things like ad-supported YouTube videos (1:20:16, l3198451)

Both platforms block prompts naming real singers, music directors, or labels, living or deceased, to avoid personality-rights lawsuits (1:34:15, l3217870)

Downloading "stems" (separated vocal/instrument layers) is the workaround for frame-level editing, since neither tool supports in-app timestamped edits like removing an instrument from second 45 to 60 (1:18:14, l3198451)

11 Labs Music's persona feature saves a voice/style so it can be reused consistently across new songs, similar to Suno's persona feature (1:06:08, l3198451)

2025 recap and the shift toward agentic, structured AI

The Dec 31 guest-hosted recap framed 2025 as a year of continuous upgrades bookended by DeepSeek R1's low-cost reasoning breakthrough in January and Gemini 3's benchmark-leading release in December. It argued 2026 will pivot from novelty creative generation toward agentic tools that call other tools, produce structured/predictable outputs, and complete real-world multi-step tasks.

2025 ran from DeepSeek R1's disruptive low-cost reasoning model (Jan 20) to Gemini 3's benchmark-leading release, with every major lab shipping at least one significant update (0:12:58, l3237330)

NotebookLM notebooks can now be added as a grounding source directly inside Gemini, letting Gemini build on synthesized research instead of raw web search (0:16:59, l3237330)

MiniMax 2.1, an open-source agent model, performs multi-step tool-calling tasks such as research plus drafting plus image generation, and can run locally for full data privacy (0:41:20, l3237330)

OpenAI's Atlas browser defends against prompt-injection attacks using an attacker-vs-defender LLM loop that iterates and self-strengthens over many rounds (0:33:14, l3237330)

Predicted 2026 theme is agentic AI moving from creative novelty to structured, real-world task completion, illustrated by Sunday Robotics' Act-1 model folding laundry via imitation learning (1:26:07, l3237330)

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.

01All-in-one AI creative studiosInstead of juggling separate tabs for image, voice, and video generation, tools like Hedra bundle character…›

Hedra bundles image generation, 11Labs-powered voice, and video/avatar creation into one "studio" workflow instead of three separate tools (0:42:52, l3151699)

Users can doodle a rough sketch and convert it into a styled image or video via Hedra's sketch-to-image feature (0:38:48, l3151699)

Hedra supports character consistency by saving a generated avatar as a reusable character ID reused across clips, up to roughly two stitched clips (1:01:06, l3151699)

Notion's AI agent can create, populate, and edit databases, Gantt charts, and calendars from one prompt, drawing on GPT, Claude Opus, and Gemini 3 models on the backend (0:56:25, l3131663)

Notion AI can be asked to "ruthlessly critique" existing pages to surface content or workflow gaps rather than just generate new content (1:06:32, l3131663)

02AI music generation and commercial-use gray areasSuno and 11 Labs Music both turn a short text prompt into a complete song (auto-written lyrics plus instrum…›

Suno and 11 Labs Music generate full tracks, including auto-written lyrics and instrumentation, from a single text prompt in roughly 5-10 seconds (0:53:59, l3198451; 0:53:48, l3217870)

Commercial or monetized use of AI-generated music requires a paid plan; free-tier tracks cannot legally be monetized even for things like ad-supported YouTube videos (1:20:16, l3198451)

Both platforms block prompts naming real singers, music directors, or labels, living or deceased, to avoid personality-rights lawsuits (1:34:15, l3217870)

11 Labs Music's persona feature saves a voice/style so it can be reused consistently across new songs, similar to Suno's persona feature (1:06:08, l3198451)

032025 recap and the shift toward agentic, structured AIThe Dec 31 guest-hosted recap framed 2025 as a year of continuous upgrades bookended by DeepSeek R1's low-c…›

2025 ran from DeepSeek R1's disruptive low-cost reasoning model (Jan 20) to Gemini 3's benchmark-leading release, with every major lab shipping at least one significant update (0:12:58, l3237330)

NotebookLM notebooks can now be added as a grounding source directly inside Gemini, letting Gemini build on synthesized research instead of raw web search (0:16:59, l3237330)

MiniMax 2.1, an open-source agent model, performs multi-step tool-calling tasks such as research plus drafting plus image generation, and can run locally for full data privacy (0:41:20, l3237330)

OpenAI's Atlas browser defends against prompt-injection attacks using an attacker-vs-defender LLM loop that iterates and self-strengthens over many rounds (0:33:14, l3237330)

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
Hidra / Heatra / Hedra (spelled inconsistently across the session)	Hedra — the AI avatar/voice/video "studio" platform from a 2023 San Francisco startup
to know / Sono / Sunu	Suno — the AI music generation app
Beat Oven / Beethoven	BeatOven.ai — an AI music generation platform named as a Suno/11 Music rival
Kria	Krea — an AI image generation platform mentioned for cheap/lower-quality side-by-side comparisons
Formally / Formerly	An unnamed Google DeepMind marketing-campaign generation platform for marketers; exact product name uncertain from audio
IndicLLM arena / in link LM	India's IndicLLM Arena — a tool for blind side-by-side comparison of LLMs on Indian-language prompts

True on recording day — verify before relying

T
h
i
s
d
i
g
e
s
t
s
u
m
m
a
r
i
z
e
s
l
i
v
e
w
e
e
k
l
y
w
e
b
i
n
a
r
r
e
c
o
r
d
i
n
g
s
f
r
o
m
D
e
c
e
m
b
e
r
2
0
2
5
;
t
o
o
l
n
a
m
e
s
,
p
r
i
c
i
n
g
,
c
r
e
d
i
t
c
o
s
t
s
,
f
e
a
t
u
r
e
a
v
a
i
l
a
b
i
l
i
t
y
,
a
n
d
c
o
u
n
t
r
y
r
o
l
l
o
u
t
s
(
e
.
g
.
H
e
d
r
a
,
S
u
n
o
,
1
1
L
a
b
s
M
u
s
i
c
,
N
o
t
i
o
n
A
I
,
G
e
m
i
n
i
A
I
m
o
d
e
f
e
a
t
u
r
e
s
)
c
h
a
n
g
e
f
r
e
q
u
e
n
t
l
y
a
n
d
s
h
o
u
l
d
b
e
r
e
v
e
r
i
f
i
e
d
a
g
a
i
n
s
t
c
u
r
r
e
n
t
v
e
n
d
o
r
d
o
c
u
m
e
n
t
a
t
i
o
n
b
e
f
o
r
e
r
e
l
y
i
n
g
o
n
a
n
y
s
p
e
c
i
f
i
c
c
l
a
i
m
,
p
r
i
c
e
,
o
r
c
a
p
a
b
i
l
i
t
y
.
S
e
v
e
r
a
l
p
r
o
d
u
c
t
n
a
m
e
s
w
e
r
e
t
r
a
n
s
c
r
i
b
e
d
i
n
c
o
n
s
i
s
t
e
n
t
l
y
b
y
A
S
R
(
s
e
e
a
s
r
C
o
r
r
e
c
t
i
o
n
s
)
a
n
d
a
f
e
w
s
p
e
a
k
e
r
c
l
a
i
m
s
(
e
.
g
.
e
x
a
c
t
c
r
e
d
i
t
c
o
s
t
s
,
p
l
a
n
l
i
m
i
t
s
,
l
e
g
a
l
/
c
o
p
y
r
i
g
h
t
i
n
t
e
r
p
r
e
t
a
t
i
o
n
s
)
a
r
e
t
h
e
p
r
e
s
e
n
t
e
r
'
s
i
n
f
o
r
m
a
l
u
n
d
e
r
s
t
a
n
d
i
n
g
a
s
o
f
l
a
t
e
D
e
c
e
m
b
e
r
2
0
2
5
,
n
o
t
v
e
r
i
f
i
e
d
l
e
g
a
l
o
r
v
e
n
d
o
r
g
u
i
d
a
n
c
e
.