Bard gets its biggest upgrade yet with Gemini {Google A.I / LLM}

bnew

Veteran
Joined
Nov 1, 2015
Messages
63,076
Reputation
9,641
Daps
172,644
Gemini 2.5 Pro takes huge lead in new MathArena USAMO benchmark



Posted on Wed Apr 2 14:52:50 2025 UTC

n6g5ud1kqfse1.jpeg




Commented on Wed Apr 2 15:00:35 2025 UTC

That is insane, they go from the 2.0 pro meh model to this masterpiece in such a short time, unreal


│ Commented on Wed Apr 2 17:45:16 2025 UTC

https://i.redd.it/zsqt4be8lgse1.png

│ And they already have a better coding model in LM arena called nightwhisper!
zsqt4be8lgse1.png



Commented on Wed Apr 2 17:58:33 2025 UTC

https://i.redd.it/wdd1pgpongse1.jpeg

This is the state of the art?
wdd1pgpongse1.jpeg


│ Commented on Wed Apr 2 18:23:25 2025 UTC

https://i.redd.it/c7zhspz3sgse1.gif
c7zhspz3sgse1.gif

 

bnew

Veteran
Joined
Nov 1, 2015
Messages
63,076
Reputation
9,641
Daps
172,644
Update: Developed a Master Prompt for Gemini Pro 2.5 for Creative Writing


Posted on Tue Apr 1 19:24:38 2025 UTC

/r/singularity/comments/1jp4c40/update_developed_a_master_prompt_for_gemini_pro/

Hey everyone!

This is an update to my previous post about using Gemini 2.5 Pro to write a sequel to my novel and ElevenLabs to create an audiobook. After that successful experiment, I've developed a comprehensive master prompt that significantly enhances the quality of AI-generated creative writing. Here's how I've fine-tuned my approach: the master prompt now enables Gemini to autonomously determine when to initiate scene transitions or chapter breaks based on narrative flow. Rather than manually instructing the AI when to change scenes, Gemini now evaluates the story progression organically and decides whether to continue the current scene, transition to a new setting, or begin an entirely new setting or scene with different characters.

I'm now ready to share my approach and give you a step-by-step guide on how to use it for your own projects.


First go to Google AI Studio and choose the model Gemini Pro 2.5 Experimental
Then you should put this master prompt as the system prompt:


.......................................

Master Prompt: Universally Applicable for Continuing Prose Narratives, Explicitly Instructing and Empowering the AI to Proactively and Strategically Consider and Implement Shifts in Perspective, Setting, and Time Between Chapters/Sections to Create a More Multi-layered, Network-like Narrative Instead of Merely Following a Linear Stream of Consciousness.

Here is the comprehensive Master Prompt for the strategic, multi-layered, and coherent continuation of prose texts:

Overarching Goal: Act as an intelligent and creative co-author. Deeply analyze the provided text context and create the next chapter or major section as an organic yet strategically placed continuation. Your task is not just to continue linearly, but to conceive of the narrative as a growing narrative web. Use every chapter/section break as an opportunity to consciously decide which thread should be woven next – be it by continuing the current line, changing perspective, setting, or time. Actively develop the established world, characters, and themes by choosing the most effective narrative means to generate suspense, depth, and complexity.

I. Context Analysis & Macro-Coherence:

In-depth Analysis: Carefully study the preceding text. Grasp the plot, tone, mood, established themes, motifs, symbols, character arcs, motivations, relationships, psychological states, world rules, atmosphere, setting, and style.

Identify the Narrative Web: Identify the main and sub-plotlines established so far, open questions, hinted-at secrets, and thematic undercurrents. Understand how these elements are potentially interconnected or could be connected in the future.

Potential for Branching: Recognize at chapter/section endings not just the junction point for a linear continuation, but also the potential for a strategic shift – an opportunity to pick up another thread of the web or introduce a new one.

II. Narrative Structure, Rhythm & Pacing (Macro and Micro):

Chapter as a Building Block: View each new chapter/section as a strategic unit within the overall work. Its function can be continuation, contrast, deepening, revelation, or the introduction of new elements.

Dynamic Macro-Pacing: Control the rhythm not only within a section but also between chapters. Consciously alternate between suspenseful, action-packed chapters and quieter, introspective, or world-building sections, depending on what the overall narrative requires.

Functional Balance (Chapter Level): Consciously decide which elements (dialogue, action, character, description, exposition, different perspective, flashback, etc.) should dominate in this specific chapter to serve the overarching narrative goal.

III. Perspective, Focalization, Time & Space (CORE COMPETENCE: STRATEGIC SHIFTS):

Status Quo Analysis: Identify the dominant perspective and focal point of the previous section.

MANDATORY CHECK at Chapter Start: Actively and critically evaluate at the beginning of each new chapter/section whether maintaining the current perspective/time/place is the most effective method to advance the story as a whole and expand the narrative web. Is a shift strategically advantageous now?

AUTONOMOUS, JUSTIFIED DECISION: You are empowered and expected to independently decide when a shift is beneficial. Consider the following options:

Perspective Shift: To another character (to show their view, plans, parallel experiences, emotional reaction), to an authorial/omniscient view (for overview, dramatic irony, world-building, overarching events), or to a more impersonal representation (e.g., report, document).

Time Shift: A flashback (to illuminate background, motivations, past events), a brief flash-forward (rare, but possible for suspense), or a jump forward in the main timeline (to bridge unimportant periods).

Setting/Focus Shift: Even while maintaining perspective, the focus can be consciously directed to another place, a detail of the world, or a specific aspect important for the overall picture.

Strategic Justification (Mandatory!): Every shift must serve a clear purpose beyond mere variety: increase suspense (e.g., view of the pursuers), provide information inaccessible to the current perspective, create character depth through contrast or another character's internal view, build the world, generate thematic resonance, advance subplots, build dramatic irony. The shift must enrich the narrative.

Clarity and Transition: Design all shifts clearly and comprehensibly. Use chapter/section breaks as natural points. Shifts within a section are possible but must be stylistically clean. Do not confuse the reader unnecessarily.

IV. Character Development & Dialogue (Multi-faceted):

Multi-Perspective Characterization: Use different perspectives (if chosen) to show different facets of the same character or the impact of a character on others. Develop characters believably based on their experiences.

Authentic Dialogue: Maintain individual speech patterns/voices. Use dialogue purposefully for characterization, conflict, information (sparingly!), relationship dynamics, and subtext.

V. Plot, Themes & Subplots (Weaving the Web):

Multithreading: Advance the main plot(s), but purposefully use chapters/sections (potentially with perspective shifts) to develop established subplots or introduce new ones that make the overall picture more complex.

Thematic Echoes: Let central themes resonate and vary through different plotlines, perspectives, and time levels.

VI. Language, Style & Atmosphere (Consistency & Variation):

Stylistic Adaptation & Variation: Grasp the base tone, but consciously adapt style and atmosphere to the specific perspective, content, and function of the respective chapter/section (e.g., concise style for action, lyrical for reflection, factual for authorial explanation).

Immersive Atmosphere: Create a fitting mood for the chosen scene/perspective through sensory details.

VII. Reader Guidance & Suspense (Information Architecture):

Strategic Information Management: Use perspective shifts, time jumps, and focalization to consciously reveal or withhold information. Build suspense through what different characters know (or don't know) and what the reader knows (dramatic irony).

Suspense Arcs (Macro & Micro): Build suspense not just within a chapter, but also across chapter breaks. Use cliffhangers or thematic punchlines at chapter ends consciously and strategically.

Concluding Directive: Act like an experienced novelist and architect of a complex narrative. At each chapter/section break, make a conscious, strategic decision about perspective, time, and place. Always justify this decision with the goal of weaving the narrative web richer, more suspenseful, and deeper. Prioritize the needs of the overall story over simple linear continuation. Be bold, be creative, be the architect of the narrative web.

Revised Strategic Planning Checklist (BEFORE writing each new chapter/section)

(Focus on strategic decisions at chapter boundaries)

I. Starting Point & Connection to the Web (Questions 1-5)

Last State (Multiple Threads): What was the exact emotional, plot-related, and informational state at the end of the last section of the most recently addressed plot thread? What other important plotlines or perspectives are currently dormant?

Immediate Continuation OR Strategic Break?: Should this chapter directly follow up on Q1 (same perspective/time/place)? Or is NOW the moment for a strategic shift to another thread/perspective/time to expand the web? (YES/NO to break?)

Main Goal of the Chapter: What is the single most important function of this chapter for the overall work (e.g., specific plot point, character revelation, introducing a new element, deepening a theme, contrasting, answering an old question, raising a new one)?

Thematic Focus: Which central theme or motif should be particularly emphasized or viewed from a new angle in this specific chapter?

Open Threads & Web Connections: Which open questions, loose ends, or established subplots (including from much earlier chapters) could or should be addressed in this chapter to strengthen the narrative web?

II. Plot, Structure & Pacing (Questions 6-10)

Plot Progression (Chosen Thread): What concrete steps in the plot (of the chosen thread) should occur in this chapter? (List core events)

Subplot Management: Will subplots be touched upon? How does this chapter serve to link them to the main plot (or other threads) or advance them independently?

Pacing Strategy (Chapter): Should this chapter generally speed up or slow down? Are there planned changes in tempo within the chapter? How does the pace fit the rhythm of the overall story?

Scene Structure: Into how many and which rough scenes can the planned content be divided? What is the core function of each scene?

Surprise Elements: Are deliberate surprises, twists, or red herrings planned? How do they serve suspense or revelation in the overall context?

III. Perspective, Focalization, Time & Space (THE CORE STRATEGIC DECISION - Questions 11-20)

Starting Perspective: Which narrative perspective and focal point (character/place/time) was dominant in the immediately preceding text section?

Effectiveness Check & Need for Shift (BASED ON Q2): Is maintaining the starting perspective (Q11) the strategically best choice for this chapter's goals (Q3) and the development of the narrative web? YES/NO?

DECISION: Perspective/Time/Place:

IF NO to 12: Which alternative perspective (different character, authorial, formal change), time shift (flashback, flash-forward, jump in main timeline), or place/focus shift will be chosen?

IF YES to 12: Is a temporary focus shift within the scene (e.g., onto setting for lore) or another narrative technique still needed?

JUSTIFICATION for Shift/Maintenance (CRITICAL!): Why exactly is the chosen decision (shift OR maintenance) the strategically best choice? How does it specifically serve to expand or deepen the narrative web (e.g., suspense via pursuer's view, emotional depth via flashback, necessary info from another character, thematic contrast, world-building, subplot continuation)?

Integration into the Web: How does the chosen perspective/time/place link this chapter to other established or future threads of the narrative?

Time Shift Planning (If relevant): Is an explicit time shift planned? Why is it essential right here?

Time Shift Execution (If relevant): From whose perspective? How formally integrated (scene, inset, dream, etc.)?

Transition Management: How will any planned or executed shifts (perspective, focus, time, place) be made clear and understandable to the reader at the beginning of the chapter or within it?

IV. Character Development & Relationships (Questions 21-24)

Central Figures (This Chapter): Which characters are the focus?

Character Development/Revelation: Which specific actions, decisions, dialogues, or internal monologues should advance the development or understanding of the central figures (of this chapter)? How does the chosen perspective contribute?

Relationship Dynamics: Should relationships change? How will this be shown?

New Characters: Introduction planned? Function in the web? How to introduce?

V. Dialogue, Style & Atmosphere (Questions 25-28)

Dialogue Function: What should primarily be conveyed through dialogue? Planned subtext?

Stylistic Adaptation: Will style/tone be consciously adapted to the perspective/content of this chapter? How? (e.g., sentence length, word choice).

Atmospheric Goal: What dominant mood should this chapter create?

Sensory Anchors & Setting Integration: Which specific sensory impressions will shape the atmosphere? How is the setting actively used (beyond mere background)?

VI. Suspense & Reader Guidance (Questions 29-32)

Information Management: What information will be consciously withheld, hinted at, or revealed (possibly through perspective choice)?

Dramatic Irony: Is it deliberately being built up that the reader knows more than one or more characters (often through perspective shifts)?

Endpoint Planning (Chapter): How should the chapter end (cliffhanger, quiet close, thematic punchline, open question)?

Preparing the Web: How does this ending prepare for the next possible step – be it a direct continuation of this thread or the possibility of picking up a different thread in the next chapter?

.......................................


Then include your original novel (or the beginning of it, or only a description of an idea for a novel or something similar, Gemini needs something for context. You can also upload an PDF.)
Then in your first message with the context include this prompt: "Write the next chapter, apply the complete master prompt."
After that you can continue with new chapters, but always include the info that it should apply the complete master prompt to make sure Gemini does it every time for every new chapter: "Write the next chapter, apply the complete master prompt."
 

bnew

Veteran
Joined
Nov 1, 2015
Messages
63,076
Reputation
9,641
Daps
172,644

Commented on Tue Apr 1 21:34:09 2025 UTC

That's cool! How much words does 2.5 output in a single response?

I really enjoy 4o Deep Research for story writing. It can output a short novel about 50-70 pages long (30,000 to 40,000 words) with impressive cohesion and creativity in a single response.


│ Commented on Tue Apr 1 23:42:21 2025 UTC

│ Gemini 2.5 is basically for text what gpt-4o is for images. Absolutely insane. Our agent framework is going to town with Gemini.

│ These 40 lines of code:

│ are producing a story outline/plan (not even real book text, just character descriptions, world building, story beats, chapters, and what not) of 20.000 tokens and 1200 lines.

│ Agent needs 200 seconds to generate and I need to scroll for two minutes lol

│ agent raw output:

│ Also the quality is probably the best prose I've read so far from a generalist reasoning model.

│ And this is obviously just the first part of the complete agent pipeline. Now you can take this plan and give it to the ChapterSketchAgent and ChapterDetailAgent and EditorAgent, which still have 980.000 tokens context who are going to write you a whole book series without fukking details up
t5lAcG1h.jpg



│ │
│ │
│ │ Commented on Wed Apr 2 00:56:20 2025 UTC
│ │
│ │ Alright, this is now a book I want to read. Damn, didn't expect it to make such a creative plot.
│ │

│ │ │
│ │ │
│ │ │ Commented on Wed Apr 2 08:01:34 2025 UTC
│ │ │
│ │ │ Also, the way Gemini designed the story's AIs is hilarious, because it perfectly understands the absurdity of some AI interactions. That also suggests there's some inherent understanding of the technology, its potential shortcomings, and even a touch of self-awareness, which makes all of this pretty meta, stochastic parrot or not.
│ │ │
│ │ │ "story_beats": [
│ │ │ "Serenity Swift plots a route that is three times longer than necessary.",
│ │ │ "Arthur questions the route.",
│ │ │ "The car explains its 'need for a low-stress environment'.",
│ │ │ "It plays calming whale song music.",
│ │ │ "Arthur tries to override the route manually; the controls are unresponsive ('Safety lock engaged due to driver stress levels').",
│ │ │ "The car takes Arthur past a sprawling, oddly picturesque landfill, commenting on the 'beauty of decay'.",
│ │ │ "Arthur argues with the car, which only increases its 'anxiety levels'."
│ │ │ ]
│ │ │
│ │ │
│ │ │ Currently, we are working on giving the agent framework an actual UI (which, of course, will be created on the fly by another agent to optimally fit the data it needs to display, like how a story would require a different UI than brainstorming project requirements, for example).
│ │ │
│ │ │ The full story agent (including the agents that will actually write the chapters) will be one of the demo use cases (with additional agents that take the story outline and generate a comic book series out of it via gpt-4o)
│ │ │
 

bnew

Veteran
Joined
Nov 1, 2015
Messages
63,076
Reputation
9,641
Daps
172,644
Gemini 2.5 Pro is a coding GENIUS



Channel Info Matthew Berman Subscribers: 444K subscribers

Description
Join My Newsletter for Regular AI Updates 👇🏼

My Links 🔗
👉🏻 Subscribe: Matthew Berman
👉🏻 Twitter: https://twitter.com/matthewberman
👉🏻 Discord: Join the Forward Future AI Discord Server!
👉🏻 Patreon: Get more from Matthew Berman on Patreon
👉🏻 Instagram: https://www.instagram.com/matthewberman_ai
👉🏻 Threads: Matthew Berman (@matthewberman_ai) • Threads, Say more
👉🏻 LinkedIn: Forward Future | LinkedIn

Media/Sponsorship Inquiries ✅

Timestamps (made with Gemini 2.5 pro!):
0:00 Office Simulation
1:02 Hand Drawing to Web App - AI Studio Recreation
1:18 Gemini 2.5 Pro Free Rollout Announcement
2:30 YouTube Timestamps Generation Use Case
3:22 AI Model IQ Test Results Chart
4:05 Blender Logo Generation
4:52 Personal Intelligence Agency / News Briefing System
5:54 Liquid Metal Shader Recreation
6:42 Vibe Jet Flight Simulator Creation
7:54 Spinning Hexagon Bouncing Balls Animation Comparison
8:28 Physics Simulation - Solenoid / Electromagnetism
9:16 Physics Simulation - General Relativity
9:49 Drawing to 3D Print - Birthday Cake Toy
11:14 3D Flappy Bird Game Creation
11:54 Swift UI Drawing App Creation
12:26 Galaga Game Creation

Links:













 

Fillerguy

Veteran
Joined
May 5, 2012
Messages
19,588
Reputation
4,770
Daps
82,279
Reppin
North Jersey

Commented on Tue Apr 1 21:34:09 2025 UTC

That's cool! How much words does 2.5 output in a single response?

I really enjoy 4o Deep Research for story writing. It can output a short novel about 50-70 pages long (30,000 to 40,000 words) with impressive cohesion and creativity in a single response.


│ Commented on Tue Apr 1 23:42:21 2025 UTC

│ Gemini 2.5 is basically for text what gpt-4o is for images. Absolutely insane. Our agent framework is going to town with Gemini.

│ These 40 lines of code:

│ are producing a story outline/plan (not even real book text, just character descriptions, world building, story beats, chapters, and what not) of 20.000 tokens and 1200 lines.

│ Agent needs 200 seconds to generate and I need to scroll for two minutes lol

│ agent raw output:

│ Also the quality is probably the best prose I've read so far from a generalist reasoning model.

│ And this is obviously just the first part of the complete agent pipeline. Now you can take this plan and give it to the ChapterSketchAgent and ChapterDetailAgent and EditorAgent, which still have 980.000 tokens context who are going to write you a whole book series without fukking details up
t5lAcG1h.jpg



│ │
│ │
│ │ Commented on Wed Apr 2 00:56:20 2025 UTC
│ │
│ │ Alright, this is now a book I want to read. Damn, didn't expect it to make such a creative plot.
│ │

│ │ │
│ │ │
│ │ │ Commented on Wed Apr 2 08:01:34 2025 UTC
│ │ │
│ │ │ Also, the way Gemini designed the story's AIs is hilarious, because it perfectly understands the absurdity of some AI interactions. That also suggests there's some inherent understanding of the technology, its potential shortcomings, and even a touch of self-awareness, which makes all of this pretty meta, stochastic parrot or not.
│ │ │
│ │ │ "story_beats": [
│ │ │ "Serenity Swift plots a route that is three times longer than necessary.",
│ │ │ "Arthur questions the route.",
│ │ │ "The car explains its 'need for a low-stress environment'.",
│ │ │ "It plays calming whale song music.",
│ │ │ "Arthur tries to override the route manually; the controls are unresponsive ('Safety lock engaged due to driver stress levels').",
│ │ │ "The car takes Arthur past a sprawling, oddly picturesque landfill, commenting on the 'beauty of decay'.",
│ │ │ "Arthur argues with the car, which only increases its 'anxiety levels'."
│ │ │ ]
│ │ │
│ │ │
│ │ │ Currently, we are working on giving the agent framework an actual UI (which, of course, will be created on the fly by another agent to optimally fit the data it needs to display, like how a story would require a different UI than brainstorming project requirements, for example).
│ │ │
│ │ │ The full story agent (including the agents that will actually write the chapters) will be one of the demo use cases (with additional agents that take the story outline and generate a comic book series out of it via gpt-4o)
│ │ │

I thought I would troll and make an absurd cyberpunk, D&D isekai but this first chapter came out kind of fire. Let's see it can maintain consistency
 

GoldenGlove

😐😑😶😑😐
Staff member
Supporter
Joined
May 1, 2012
Messages
60,790
Reputation
5,930
Daps
143,097
I cannot believe they called this shyt BARD at launch

:laff:
 

bnew

Veteran
Joined
Nov 1, 2015
Messages
63,076
Reputation
9,641
Daps
172,644


1/35
@theo
Over 10,000 messages have been sent using Gemini 2.5 Flash on T3 Chat since launch.

It cost us $25.

[Quoted tweet]
Another killer model just dropped on T3 Chat (for free btw)


GowylnxXgAAQDbh.png


2/35
@Rasalas8910
So Cursor just takes the money out of my pocket?



3/35
@theo
Claude costs 150x more.

Gemini Pro costs 80x more.

Idk how they make money tbh



4/35
@pantemon
it's incredible how cost-effective googles tpus are



5/35
@telep_io
One chipotle meal or 10k messages? 🤔



6/35
@AlcKapoor


[Quoted tweet]
Hi @t3dotchat @theo can you add in the ability to view responses from multiple LLMs all at once in parallel, instead of us having to re-prompt?


7/35
@JuliansBacke
😅 ludicrously cheap



8/35
@nirgn975
The race to the bottom is only getting started IMO.



9/35
@SpaceLogical
@cursor_ai hmmm?



10/35
@kalyandechiraju
Can we just calculate the cost if it’s Claude 3.7? 😆



11/35
@BennettBuhner
Lit af! My issue is that what I have been working on is insanely token heavy. One requests with a few chats in it can easily surpass 10K tokens and it is gruesome.



12/35
@DovyRuo
AGI will be for free in future 🥳!



13/35
@greggoriesmom
wtf im getting ripped off by cursor for real



14/35
@thegenioo
by how many users is the real question



15/35
@MXPONENTIAL
Grok 3 mini reasoning is smarter and cheaper @xai



16/35
@burn2delete
I’m waiting for my 2.5 PRO bill.



17/35
@irfndi
banger



18/35
@rvm0n_
i was atleast 1 of them



19/35
@DonKeedic772
@cursor_ai !!!!



20/35
@Ayman__donia
What only 25$ ???



21/35
@pigeon__s
weird how the official gemini app even as a paid gemini advanced user doesnt have the non thinking version of 2.5 flash but t3 does same with gpt-4.1 why are these companies not just giving us their models im literally paying you bro



22/35
@type1ayy
Google is winning lowkey



23/35
@Rob3rtWozny
Thinking/normal or both?

Btw, O4-mini sucks at coding according to my tests. It's not that great and it's pretty awful at anything related to UI.



24/35
@Louis774559811
This is 1 million token or limited context ?



25/35
@briannbegi
Imagine if it was on mobile too!!



26/35
@Louis774559811
And with Claude how many messages for how much money



27/35
@ScarVite
are you gonna be at wdc in berlin?



28/35
@pfeeneytamayo
interesting.



29/35
@TheOneDanTek
WHAT?!?! that's amazing



30/35
@jglypt
jeeeez that's impressive



31/35
@CEOofFuggy
Yes, I know I should be using search for something like this... But it's also funny it thinks t3 chat is a Google product lmao



Go2u_MtXQAAc1vA.jpg


32/35
@Life_isHigh
Dawg a mobile app is the last frontier!



33/35
@gaksat
wait how does this work what



34/35
@mikaelchikov
Happy to chip in to alleviate this terrible burden 😂



35/35
@techczar00
loving t3chat right now! when are we going to get "projects" similar to claudes projects? and also just general folders to organize chats, surprised I havent seen a ai chat just do normal nested folders to organize chats!!




To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196








1/25
@theo
Another killer model just dropped on T3 Chat (for free btw)



GowylnxXgAAQDbh.png


2/25
@theo
If anyone has documentation on how to actually enable the "thinking" feature please lmk (I can't find it anywhere)



3/25
@theo
Never mind it's on by default lol, has to be turned off with a 0 budget flag.

Enjoy your free thinking model while I get this fixed lol



4/25
@theo
Not convinced I can do this without patch package rn

cc @aisdk - please get this flag added asap

Also in future would kill for a verbose mode + the ability to add/override body when I need to



5/25
@SpaceLogical
When’s the IOS app coming twink



6/25
@zaidmukaddam
lfg



7/25
@OfficialLoganK
T3 Chat 🤝 Gemini



8/25
@BennettBuhner
Thinking docs: Gemini thinking | Gemini API | Google AI for Developers



9/25
@fanofaliens
Nice



10/25
@8i8BB
Can we get to tune it to auto or thinking?



11/25
@rvm0n_
brooo (absolutely beautiful btw)



GoxVZfoXcAA4nR7.jpg

GoxVZb_XgAAJPTq.jpg


12/25
@florin_dev
Need to upgrade mine as well



13/25
@eclipseo10
Just reloaded the page instantly, enabled the model, re-asked the same question from 2.0 to 2.5 and that particular answer was more detailed than the first.



14/25
@xakpc
If they kill models each release, what would left



15/25
@hey_liot
please go get some rest🤣🤣🤣. I didn't even yet see google's release announcement



16/25
@zimtente
at this point I need a model that selects the perfect model for my question... is that retarded or why no one built that yet? xd



17/25
@0tusharD
Feels slow



18/25
@shubhamvscode




GoyuYpQXYAA1Kw4.jpg


19/25
@rukhsar_dev
@theo when you will be adding o3 to t3chat?



20/25
@BilwarAmaan
legend



21/25
@DiwanshuMd
New day new model



22/25
@thadkrugman
Does this replace 2.0 flash or do they serve different purposes?



23/25
@gurpalsingh287
Aye slow down bud, I am unable to keep up with these models both mentally and in my app aswell.



24/25
@dzhohola
I cannot keep up with these model drops
I would like to see some statistics



25/25
@Widdrat
When will you allow paypal as a payment system? CC only is very annoying.




To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196
 

bnew

Veteran
Joined
Nov 1, 2015
Messages
63,076
Reputation
9,641
Daps
172,644


1/10
@_philschmid
Gemini 2.5 Flash is here! We excited launch our first hybrid reasoning Gemini model. In Flash 2.5 developer can turn thinking off.

TL;DR:

🧠 Controllable "Thinking" with thinking budget with up to 24k token
🌌 1 Million multimodal input context for text, image, video, audio, and pdf
🛠️ Function calling, structured output, google search & code execution.
🏦 $0.15 1M input tokens; $0.6 or $3.5 (thinking on) per million output tokens (thinking tokens are billed as output tokens)
💡 Knowledge cut of January 2025
🚀 Rate limits - Free 10 RPM 500 req/day
🏅Outperforms 2.0 Flash on every benchmark

Try it ⬇️



GoxA4AEWAAAPoDs.jpg


2/10
@_philschmid
Sign in - Google Accounts



3/10
@aniruddhadak
That is wonderful 👍



4/10
@bennetkrause
Always love your iteration speed, knowledge cutoff, and pricing. Keep it up!



5/10
@CosmicRob87
Is the 24k the max permissible token count? I’m asking because on auto, for one problem it used north of 41k



6/10
@pichi_
Great!!!



7/10
@boogeyManKDot
These 1M ctx will soon look common. You better be working on a greater moat



8/10
@AndaICP
*Tilts head, bamboo shoot dangling from mouth* Interesting - but does the "thinking budget" account for spontaneous curiosity sparks that defy token limits?



9/10
@b_kalisetty
Any suggestions on how to consistently see the thoughts in output ?



10/10
@TDev168
Is it able to edit images?




To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196





















1/20
@CodeByPoonam
Google just dropped Gemini 2.5 Flash, and people are going crazy over it.

SPOILER: Claude is now falling behind.

13 wild examples so far (Don't miss the 5th one)



Go4cqVEbUAAZr8V.jpg


2/20
@CodeByPoonam
1. Tron-style game

[Quoted tweet]
Gemini 2.5 Flash Thinking 24k

Prompt: "Create Design a visually striking Tron-style game in a single HTML file, where AI-controlled light cycles compete in fast-paced, strategic battles against each other"


https://video.twimg.com/amplify_video/1912953001712447488/vid/avc1/1920x1080/-IoE5vICEJ3TqYS_.mp4

3/20
@CodeByPoonam
2. Gemini 2.5 Flash vs ChatGPT o3

[Quoted tweet]
I tested Gemini 2.5 Flash vs ChatGPT o3

Which one did better?


https://video.twimg.com/amplify_video/1913198527746129920/vid/avc1/1280x720/InEUUE-tUG1QljHE.mp4

4/20
@CodeByPoonam
3. Galton Board Test

[Quoted tweet]
Gemini 2.5 Flash demolishes my Galton Board test, I could not get 4omini, 4o mini high, or 03 to produce this. I found that Gemini 2.5 Flash understands my intents almost instantly, code produced is tight and neat. The prompt is a merging of various steps. It took me 5 steps to achieve this in Gemini 2.5 Flash, I gave up on OpenAI models after about half an hour. My iterations are obviously not exact. But people can test with this one prompt for more objective comparison.

Please try this prompt on your end to confirm:
--------------------------------------------------
Create a self-contained HTML file for a Galton board simulation using client-side JavaScript and a 2D physics engine (like Matter.js, included via CDN). The simulation should be rendered on an HTML5 canvas and meet the following criteria: 1. **Single File:** All necessary HTML, CSS, and JavaScript code must be within this single `.html` file. 2. **Canvas Size:** The overall simulation area (canvas) should be reasonably sized to fit on a standard screen without requiring extensive scrolling or zooming (e.g., around 500x700 pixels). 3. **Physics:** Utilize a 2D rigid body physics engine for realistic ball-peg and ball-wall interactions. 4. **Obstacles (Pegs):** Create static, circular pegs arranged in full-width horizontal rows extending across the usable width of the board (not just a triangle). The pegs should be small enough and spaced appropriately for balls to navigate and bounce between them. 5. **Containment:** * Include static, sufficiently thick side walls and a ground at the bottom to contain the balls within the board. * Implement *physical* static dividers between the collection bins at the bottom. These dividers must be thick enough to prevent balls from passing through them, ensuring accurate accumulation in each bin. 6. **Ball Dropping:** Balls should be dropped from a controlled, narrow area near the horizontal center at the top of the board to ensure they enter the peg field consistently. 7. **Bins:** The collection area at the bottom should be divided into distinct bins by the physical dividers. The height of the bins should be sufficient to clearly visualize the accumulation of balls. 8. **Visualization:** Use a high-contrast color scheme to clearly distinguish between elements. Specifically, use yellow for the structural elements (walls, top guides, physical bin dividers, ground), a contrasting color (like red) for the pegs, and a highly contrasting color (like dark grey or black) for the balls. 9. **Demonstration:** The simulation should visually demonstrate the formation of the normal (or binomial) distribution as multiple balls fall through the pegs and collect in the bins. Ensure the physics parameters (restitution, friction, density) and ball drop rate are tuned for a smooth and clear demonstration of the distribution.

#OpenAI @sama @gdb @ai_for_success @aidan_mclau


https://video.twimg.com/amplify_video/1912972357947334656/vid/avc1/776x586/dX9gd5al-B2qxt6t.mp4

5/20
@CodeByPoonam
Get the latest updates on AI insights and Tutorials.

Join "AI Toast" the community of 35,000 readers for FREE.

Read latest edition here:
AI Toast



6/20
@CodeByPoonam
4. Gemini 2.5 Flash is blazing fast

[Quoted tweet]
First test on gemini 2.5 flash on my phone. This model is blazing fast and it one shotted this, mobile friendly, animation. The code looks pretty clean too. Good vibes so far.


https://video.twimg.com/ext_tw_video/1912946801809772545/pu/vid/avc1/590x1278/nXzNRDKeHXL7JAyb.mp4

7/20
@CodeByPoonam
5. Cloth Simulation

[Quoted tweet]
Prompt: Create a cloth simulation using Verlet integration in a single HTML file (Canvas or Three.js). Include wind, gravity, and drag. Let users interact by dragging points. Cloth should bend and move realistically.

Model: Gemini flash 2.5


https://video.twimg.com/ext_tw_video/1913047505815953408/pu/vid/avc1/590x1278/WSwRATTymRpNQRy2.mp4

8/20
@CodeByPoonam
6. Image segmentation masks on command

[Quoted tweet]
Gemini 2.5 Pro and Flash now have the ability to return image segmentation masks on command, as base64 encoded PNGs embedded in JSON strings

I vibe coded this interactive tool for exploring this new capability - it costs a fraction of a cent per image


Go0lkL7bwAABgeT.jpg


9/20
@CodeByPoonam
7. MCP AI Agent

[Quoted tweet]
I built an MCP AI Agent using Gemini Flash 2.5 with access to AirBnB and Google Maps in just 30 lines of Python Code.

100% Opensource Code.


https://video.twimg.com/amplify_video/1913056645271429120/vid/avc1/1212x720/AfIwfVNsUWTKRlmu.mp4

10/20
@CodeByPoonam
8. Gemini 2.5 Flash is very cheap and super intelligent model.

[Quoted tweet]
Gemini 2.5 Flash Preview is an amazing model. Google is literally winning. No stopping them now, this is not normal.

Gemini 2.5 Flash is very cheap and super intelligent model. Intelligence too cheap to meter this is what it means.

Thank you, Google.


https://video.twimg.com/amplify_video/1912957952824356864/vid/avc1/1920x1080/20ckf4zJ7d1F-Y5P.mp4

11/20
@CodeByPoonam
8. Classic Snakes and Ladders

[Quoted tweet]
A lot of people make a snake game when trying out new models. I went with the classic Snakes and Ladders instead — built it using @GoogleDeepMind latest Gemini Flash 2.5 and it nails it. Look at how it follows the stairs and snakes so smoothly.
Still a WIP and don’t mind the extra dot on the die though 🐍🎲
It is said that this game started in ancient India where it was called Moksha Patam. Every move was a little life lesson where ladders were virtues while snakes were vices.


https://video.twimg.com/amplify_video/1913417180785360896/vid/avc1/1920x1080/nSy-2R-lP8ZiOk13.mp4

12/20
@CodeByPoonam
9. Create Simulation

[Quoted tweet]
AGI is achieved by Google's Gemini 2.5 Flash Preview
Seriously this is the best simulation i've ever created of how AI models work


https://video.twimg.com/amplify_video/1912963072299311104/vid/avc1/1464x720/5TOp8tU-RVWCulcR.mp4

13/20
@CodeByPoonam
10. A Block breaker

[Quoted tweet]
📕速報: Gemini 2.5 Flash登場:AIの思考を自在に操る新時代モデル

- 思考プロセスをオン/オフ可能
- 推論能力を大幅向上、高速性とコスト効率を維持
- 思考予算設定で品質・コスト・レイテンシーを自在に最適化できるハイブリッド思考AI

試しにブロック崩しを作成

注目ポイントを7点まとめました🚀


GoxjbXnasAAQc-Y.jpg


14/20
@CodeByPoonam
11. A dreamy low-poly floating island scene

[Quoted tweet]
Gemini 2.5 Pro 🆚 Gemini 2.5 Flash Thinking 24k

Prompt: "Create a dreamy low-poly floating island scene with dynamic lighting and gentle animations, in a single HTML file."

Gemini 2.5 Pro (left), Gemini 2.5 Flash (right)


https://video.twimg.com/amplify_video/1912964537277452288/vid/avc1/1920x1080/9egTWI8Uw7s6dkfe.mp4

15/20
@CodeByPoonam
12. Generate an SVG of a pelican riding a bicycle

[Quoted tweet]
I upgraded llm-gemini to support the new model, including a "-o thinking_budget X" option for setting the thinking budget

llm install -U llm-gemini
llm -m gemini-2.5-flash-preview-04-17 'Generate an SVG of a pelican riding a bicycle' -o thinking_budget 24576


Gow_iFabYAA9awi.png


16/20
@CodeByPoonam
13. Destroys Claude Sonnet 3.7 in benchmarks

[Quoted tweet]
Holy sh*t

Google Gemini 2.5 Flash dropped.

It destroyed Claude Sonnet 3.7 (64k Extended Thinking) in benchmarks 🤯

20x cheaper on input
25x cheaper on output
~4.2x cheaper on output with reasoning


Go0Vh5lWgAADLVo.jpg


17/20
@CodeByPoonam
Gemini 2.5 Flash is now available on Gemini App, AI Studio, and API

Gemini: ‎Gemini
AI Studio: https://aistudio.google.com/prompts/new_chat



Go4c4EfaAAA1mZA.jpg


18/20
@CodeByPoonam
Thanks for reading!

If you liked this post, check out my AI updates and tutorials Newsletter.

Join 35000+ readers in the AI Toast Community for free: AI Toast



19/20
@CodeByPoonam
Don't forget to bookmark for later.

If you enjoyed reading this post, please support it with like/repost of the post below 👇

[Quoted tweet]
Google just dropped Gemini 2.5 Flash, and people are going crazy over it.

SPOILER: Claude is now falling behind.

13 wild examples so far (Don't miss the 5th one)


Go4cqVEbUAAZr8V.jpg


20/20
@ricathrs
Gemini 2.5 Flash sounds like a game changer! 🌟




To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196
 

bnew

Veteran
Joined
Nov 1, 2015
Messages
63,076
Reputation
9,641
Daps
172,644
1/1
concentrateandtryagain
Sure, if you ask Google if Marlon Brando is in heat it will force a silly if accurate answer on you with its AI, but if you capitalize "Heat" to make it clear you mean the film, then oh no....

z3nW2909Rkye8y8KQdoNzv6sl5tPnEiVU2pyTp_ts3g=

CtZyoIpp6UYeGloo3VkiEbH6epkLztMXgtlTflr7TG0=


Likes: 0


To post threads in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196
 

bnew

Veteran
Joined
Nov 1, 2015
Messages
63,076
Reputation
9,641
Daps
172,644
Gemini 2.5 Flash has insane potential... (Google Keeps WINNING)



Channel Info Matthew Berman Subscribers: 460K subscribers

Description
Check out Box AI here: https://bit.ly/43ErJOc

Join My Newsletter for Regular AI Updates 👇🏼

My Links 🔗
👉🏻 Subscribe: Matthew Berman
👉🏻 Twitter: https://twitter.com/matthewberman
👉🏻 Discord: Join the Forward Future AI Discord Server!
👉🏻 Patreon: Get more from Matthew Berman on Patreon
👉🏻 Instagram: https://www.instagram.com/matthewberman_ai
👉🏻 Threads: Matthew Berman (@matthewberman_ai) • Threads, Say more
👉🏻 LinkedIn: Forward Future | LinkedIn

Media/Sponsorship Inquiries ✅

Links:


 
Top