Claude 3 — THE END OF CHATGPT IS COMING. (Shocking AI News)

AI Master
9 Mar 202420:01

Summary

TLDRThe AI landscape saw significant developments with the announcement of Anthropic's CLAW 3, boasting three versions with varying power levels, and Stability AI's Stable Diffusion 3, promising advanced image generation. Microsoft introduced Co-Pilot for Finance, a tool for financial professionals, while Mistol Large from a French startup aims to compete with industry giants. Adobe previewed Music Gen AI Control for custom audio creation, and Amazon's Rufus, an AI shopping assistant, faced mixed reviews. Wix launched an AI Builder for website creation, and OpenAI shared insights into its relationship with Elon Musk and Tesla.

Takeaways

  • 🚀 Anthropic announces CLAW 3, a new AI model claimed to be more powerful than ChatGPT.
  • 📈 CLAW 3 comes in three versions: Haiku, Sonet, and Opus, offering varying levels of intelligence, speed, and cost.
  • 📊 Comparative analysis shows Opus outperforms other models in AI system benchmarks, including expert knowledge and reasoning tasks.
  • 🌐 CLAW 3 models demonstrate increased capabilities in analysis, forecasting, content creation, and non-English language understanding.
  • 🖼️ Stability AI's Stable Diffusion 3, using a multimodal diffusion Transformer architecture, promises improved image generation capabilities.
  • 🎶 Adobe previews Music Gen AI Control, a tool for generating and editing custom audio based on user prompts.
  • 💼 Microsoft's Co-Pilot for Finance aims to assist finance professionals with data analysis and routine tasks using Excel, Outlook, and Teams.
  • 🤖 Mistro's Mr. Large, a multilingual AI model, ranks high in global lists for metrics and accuracy, showing promise for European startups.
  • 🛒 Amazon's Rufus, an AI shopping assistant, is still under development and has room for improvement based on initial user feedback.
  • 📈 Wix's AI Builder simplifies website creation with AI-assisted design, though the results may be modest compared to professional design.
  • 🔄 OpenAI introduces a feedback and rating system for third-party GPT models, enhancing user satisfaction and model improvement.
  • 📝 OpenAI's blog post reveals the organization's early relationship with Elon Musk and Tesla, highlighting the importance of independent AI development.

Q & A

  • What is Anthropic's new AI model, and how does it compare to other models in terms of power and cost?

    -Anthropic announced a new AI model called CLA 3, which comes in three versions: Haiku, Sonet, and Opus. Haiku is the least powerful and Opus is the most powerful. The division into three models allows users to select the optimal balance of intelligence, speed, and cost for their application. Haiku is cheaper to run than Sonet but delivers over half the intelligence, while Opus, though more expensive, offers around 50-60% higher performance.

  • How does CLA 3 perform in comparison to other AI models from companies like Google and OpenAI?

    -CLA 3 models show increased capabilities in analysis, forecasting, content creation, and conversing in non-English languages like Spanish, Japanese, and French. Even the basic CLA 3 Haiku outperforms both GBT 3.5 and Gemini Pro, while Sonet is somewhere in the middle between GBT 3.5 and GBT 4 Gemini, Ultra, and Pro. In many tasks, it even manages to outperform the most advanced models from Google and OpenAI.

  • What improvements have been made in the vision capabilities of the latest generation of CLA models?

    -CLA 3 models, particularly the Opus version, show a significant jump in vision capabilities. Claude 3 Opus is better than GBT 4 in many ways and is very similar to Gemini Ultra in image recognition, which was proven to be really good. Additionally, the third generation of CLA makes fewer refusals and shows more nuanced understanding of requests, recognizing real harm and refusing to answer harmless prompts much less often.

  • What is the significance of the mmdi architecture used in Stability AI's Stable Diffusion 3?

    -The mmdi architecture stands for Multimodal Diffusion Transformer, which combines the strengths of both diffusion models and Transformers. Diffusion models are good at creating detail in small regions but poor at generating the overall layout of an image, while Transformers excel at layout but struggle with detail. By combining the two, Stable Diffusion 3 can leverage the best aspects of both architectures for improved image generation.

  • What are some of the limitations of Stable Diffusion 3 in text-to-image generation?

    -Stable Diffusion 3, while improved, still has issues with text generation, such as imperfections in spacing, sizing, and letter formation. For example, the letters A and B might become one letter, and there can be variable distance between letters. It also struggles with consistency in rendering shadows, as seen in images where shadows from different objects fall in different directions.

  • What is Microsoft's Co-Pilot for Finance, and how does it aim to assist finance professionals?

    -Co-Pilot for Finance is a niche product optimized for people working in finance with large spreadsheets. It is integrated into Excel, Outlook, and Teams, helping with routine tasks and data analysis. It can access and analyze data from various sources, automate repetitive tasks, generate insights and recommendations, and even suggest actions based on data.

  • How does Mr Large, the AI developed by French startup Misto, compare to other AI models like GPT-4 and Gemini Ultra?

    -Mr Large ranks second in the top LLM Global list, with metrics and accuracy only outperformed by GPT-4. It is natively fluent in English, French, Spanish, German, and Italian, with a nuanced understanding of grammar and cultural context. It has a 32k token context window, precisely follows instructions, allows for custom moderation policies, and is natively capable of function calling. In terms of multilingual capacities, Mol outperforms Meta and Lama 2 by a noticeable margin.

  • What is Adobe's new project, Music Gen AI Control, and how does it assist musicians?

    -Music Gen AI Control is a project that allows musicians to generate music from text prompts and have full control over the audio. It helps people create music for their projects, whether they are broadcasters, podcasters, or anyone else who needs audio with the right mood, tone, and length. The AI creates an initial tune based on the user's prompt, which they can edit using text commands, such as adjusting the music's intensity, extending the length of the clip, or creating a repeatable loop.

  • What are some criticisms of Amazon's Rufus AI shopping assistant?

    -Amazon's Rufus has been criticized for not living up to its promise of being an expert shopping assistant. The Washington Post's testing found that Rufus often offered basic suggestions that could be found in how-to articles and did not recommend specific products. Users also reported difficulty in understanding why certain products were suggested as the right ones for their needs, leading to a lack of trust in the AI's recommendations.

  • How does Wix's AI Builder differ from other AI website builders in terms of ease of use and customization?

    -Wix's AI Builder allows users to create a website by simply hitting the 'create with AI' button and chatting with the bot about their desired website features. The process is quick, with simple questions taking only a few seconds to answer. Users can further refine the design with additional requirements, and the AI can adjust the layout based on prompts. While the promotional materials may exaggerate its capabilities, Wix's AI Builder offers a faster and cheaper alternative to traditional website design, albeit with more modest results.

  • What is the new update to the GPT store, and how does it benefit users and creators?

    -The new update to the GPT store allows users of third-party GPTs to rate them using a drop-down menu with a 'send feedback' option. This feature enables users to directly contact the creator with suggestions for improvement or to report issues. The rating system will likely be used to indicate user satisfaction with certain GPTs, providing valuable feedback to creators and helping users make informed decisions about which GPTs to use.

  • What was the proposed relationship between OpenAI and Tesla, and why didn't it come to fruition?

    -There were discussions about making OpenAI part of Tesla, with Elon Musk suggesting that OpenAI could attach to Tesla as its 'Cash Cow' to cover the massive costs of developing AGI. However, the idea of merging with Tesla or any deal that would give someone complete control over OpenAI did not align with the organization's goals. This led to Elon Musk parting ways with OpenAI and eventually starting his own AI company.

  • What is Apple's claim about the new M3 MacBook Air in relation to AI?

    -Apple claims that the new M3 MacBook Air is the world's best consumer laptop for AI. The machine is said to be capable of running optimized AI models, including large language models and diffusion models for image generation, locally with great performance. This claim is part of Apple's broader strategy to leverage AI across all its platforms and enhance productivity and creativity.

Outlines

00:00

🤖 Introduction to CLA 3 and Stable Diffusion 3

The script begins with an overview of recent developments in AI, highlighting the announcement of CLA 3 by Anthropic, a new AI model claimed to be more powerful than Chatbot. CLA 3 is available in three versions: Haiku, Sonet, and Opus, each offering different levels of intelligence, speed, and cost-effectiveness. The script discusses the capabilities of these models, their performance in comparison to other AI systems like GPT-3.5 and Gemini variants, and their enhanced vision and comprehension abilities. It also touches on the availability of Opus and Sonet as APIs and the potential of CLA 3 in various applications. Additionally, the introduction of Stable Diffusion 3 by Stability AI is discussed, which promises significant advancements in AI-generated images, although it is not yet publicly available.

05:00

🚀 Advancements in AI for Finance and Language Models

This paragraph focuses on the announcement of Co-Pilot for Finance by Microsoft, a specialized tool designed for financial professionals to assist with spreadsheets and data analysis. It integrates with Excel, Outlook, and Teams, automating routine tasks and providing insights based on data. The script also discusses the unveiling of Mr. Large by Mistral, a French startup, which claims to be a multilingual language model similar to GPT-4. The comparison of Mr. Large with other models like GPT-4 and Gemini Ultra is highlighted, along with its capabilities in understanding grammar and cultural context. The segment concludes with a mention of Adobe's preview of Music Gen AI Control, a tool for generating and editing custom audio for various projects.

10:07

🛒 Amazon's Rufus and Wix's AI Builder Updates

The script continues with a discussion on Amazon's Rufus, an AI shopping assistant that, despite its promise, did not meet expectations during testing by the Washington Post. It then shifts focus to Wix's AI Builder, a tool that simplifies website creation by allowing users to chat with a bot about their website requirements. The AI Builder can generate a website based on user input and adjust themes and layouts accordingly. The script also notes a small update to the GPT store, allowing users to rate and provide feedback on third-party GPTs directly to their creators.

15:09

📈 OpenAI's History and Apple's M3 MacBook Air

The final paragraph delves into OpenAI's past, including its relationship with Elon Musk and Tesla, and the discussions around the organization's funding and potential for-profit models. It outlines the events that led to Musk's departure and the growth of OpenAI without his involvement. The script then mentions Apple's claim that the new M3 MacBook Air is the world's best consumer laptop for AI, highlighting its ability to run optimized AI models locally with great performance. The paragraph concludes by emphasizing the rapid pace of AI development and the continuous flow of news in the field.

Mindmap

Keywords

💡AI

Artificial Intelligence (AI) refers to the simulation of human intelligence in machines that are programmed to think and learn like humans. In the context of the video, AI is the central theme, with various advancements and new models being discussed, such as CLA 3 and stable diffusion 3, which aim to improve tasks like comprehension, content creation, and problem-solving.

💡CLA 3

CLA 3 is a new AI model announced by Anthropic, which comes in three versions: Haiku, Sonet, and Opus, each offering different levels of intelligence, speed, and cost-effectiveness. The model is designed to provide an optimal balance for various applications, with Opus being the most powerful and capable of near-human levels of comprehension on complex tasks.

💡Stable Diffusion 3

Stable Diffusion 3 is an AI model announced by Stability AI that uses a multimodal diffusion Transformer architecture, which combines the strengths of diffusion models and Transformers for improved image generation. It promises better detail creation and overall layout compared to previous versions, addressing common issues in AI image generation.

💡Finance

In the context of the video, finance refers to the application of AI in the financial sector. Microsoft's co-pilot for finance is highlighted as a niche product optimized for finance professionals, aiming to automate routine tasks, analyze data, and provide insights for more informed decision-making.

💡Mistol Large

Mistol Large is an AI model developed by a French startup, Mistro, which claims to be very similar to GBT 4 in terms of metrics and accuracy. It is multilingual, understanding grammar and cultural context in English, French, Spanish, German, and Italian, and offers capabilities like custom moderation policies and function calling.

💡Adobe Music Gen

Adobe Music Gen is a project by Adobe that introduces cutting-edge generative AI tools for crafting and editing custom audio. It allows musicians to generate music from text prompts and have full control over the audio, making it a co-creator in music production for various projects like broadcasting or podcasting.

💡Amazon Rufus

Amazon Rufus is an AI shopping assistant trained on Amazon's product catalog and web information to answer customer questions and facilitate product discovery. However, the video mentions that Rufus did not fully deliver on its promise of finding the best products or making reliable recommendations.

💡Wix AI Builder

Wix AI Builder is a feature of the Wix website building platform that allows users to create a website by simply hitting a 'create with AI' button and chatting with a bot about their desired website features. It aims to simplify the website creation process by generating a full-fledged website based on user inputs.

💡GPT Store

The GPT Store is a platform where users can find and rate third-party GPT models. It allows users to provide feedback and ratings, which can help indicate user satisfaction and guide improvements for the models.

💡Open AI and Tesla

Open AI and Tesla's relationship is discussed in the context of the video, highlighting past discussions about Tesla potentially acquiring Open AI and Elon Musk's role in its funding. The narrative explains how Open AI maintained its independence and went on to achieve significant advancements without Tesla's involvement.

💡M3 MacBook Air

The M3 MacBook Air is Apple's latest laptop that claims to be the world's best consumer laptop for AI. It is designed to run optimized AI models, including large language models and diffusion models for image generation, locally with great performance, enhancing productivity and creativity on the Mac platform.

Highlights

Anthropic announces CLAW 3, claiming it to be more powerful than ChatGPT.

CLAW 3 comes in three versions: Haiku, Sonet, and Opus, with varying levels of intelligence and cost.

Opus is the most powerful version, exhibiting near-human levels of comprehension on complex tasks.

CLAW 3 models show increased capabilities in analysis, forecasting, and content creation.

CLAW 3 has made significant improvements in vision capabilities, with Opus outperforming other models in image recognition.

Stability AI announces Stable Diffusion 3, using a multimodal diffusion Transformer architecture.

Stable Diffusion 3 focuses on improving text-to-image generation and reducing imperfections.

Microsoft's Co-Pilot for Finance is a niche product optimized for finance professionals, integrating with Excel, Outlook, and Teams.

Mistral Large, developed by a French startup, promises to be very similar to GPT-4, with multilingual capabilities and a 32k token context window.

Adobe previews Music Gen AI Control, allowing musicians to generate and edit custom audio from text prompts.

Amazon's Rufus, an expert shopping assistant, is still under development and not yet living up to its promise.

Wix releases AI Builder, simplifying website creation with AI assistance.

GPT Store gets an update, allowing users to rate and send feedback on third-party GPT models.

OpenAI releases a blog post detailing its relationship with Elon Musk and Tesla, and the discussions around funding and control.

Apple's M3 MacBook Air is marketed as the world's best consumer laptop for AI, with the ability to run optimized AI models locally.

CLAW 3 models show a nuanced understanding of requests, recognizing real harm and refusing to answer harmless prompts less often.

Stable Diffusion 3 emphasizes performance and human feedback, outperforming current text-to-image generation systems in various tests.

Mistral Large's developers are releasing a smaller, more optimized model focusing on latency and cost, faster than its competitors.

Wix's AI Builder, while promising, delivers results that are modest in creativity and advanced features, akin to a junior designer's work.

OpenAI's blog post suggests that the organization's independence from corporate control, like Elon Musk's, has been crucial to its success.

Transcripts

00:00

the last week was hot in the world of AI

00:03

claw 3 stable diffusion 3 new AI

00:06

announcements and updates open AI

00:08

acquisition and more so let's not waste

00:10

any more time and get straight to

00:12

business anthropic just announced claw

00:14

three and they claim it to be more

00:16

powerful than chbt what is it all about

00:18

the official blog post gives a lot of

00:20

information about the new model that I

00:22

think we should check out first cloth 3

00:24

comes in three different versions

00:26

depending on how smart and Powerful the

00:27

model is cloth 3 haiku Sonet and Opus

00:31

Haiku is the least powerful of the bunch

00:33

while Opus is the true Powerhouse

00:35

developers say that this division into

00:37

three models allows users to select the

00:39

optimal balance of intelligence speed

00:42

and cost for their application here is

00:44

how the graph looks we can see that claw

00:47

3 Hau is many times cheaper to run than

00:50

Sonet while delivering over half of the

00:52

intelligence and judging by this graph

00:54

the Opus is even more expensive than

00:57

Sonet while also getting around 50 to

00:59

60% % higher performance but this graph

01:02

doesn't really tell us stories so let's

01:04

look at the comparison chart of all

01:05

three claw models Chad gbt variants and

01:08

two Gemini variants Opus outperforms its

01:11

fears on most of the common evaluation

01:13

benchmarks for AI systems including

01:16

undergraduate level expert knowledge

01:19

MML graduate level expert reasoning GP

01:22

QA basic mathematics not saying that and

01:26

more it exhibits near human levels of

01:28

comprehension Flint on complex tasks

01:31

leading the frontier of general

01:33

intelligence all claw fre models show

01:36

increased capabilities and Analysis and

01:38

forecasting Nuance content creation

01:40

cenation conversing in non-english

01:43

languages like Spanish Japanese and

01:45

French what I find interesting is that

01:47

even the basic claw free Haiku

01:49

outperforms both gbt 3.5 and Gemini Pro

01:53

sonit seems to be somewhere in the

01:55

middle between gbt 3.5 and gbt 4 Gemini

01:58

Ultra and Gemini Pro and in many tasks

02:01

it even manages to outperform the most

02:03

advanced models from Google and open AI

02:06

of course all the test conducted can be

02:08

called cherry-picking and may not fully

02:10

paint a picture of how powerful the CLA

02:13

3 is another area where CLA 3 saw a

02:16

significant jump was in terms of vision

02:18

capabilities the chart shows that Claude

02:21

3 Opus in many ways is better than gbt 4

02:23

vision and in some cases is very similar

02:26

to Gemini Ultra that was proven to be

02:28

really good and image recognition also

02:31

the third generation of Claud makes

02:33

fewer refusals the Claud free models

02:35

show more Nuance understanding of

02:37

requests recognize real harm and refuse

02:40

to answer harmless prompts much less

02:43

often developers also claim to improve

02:45

the accuracy of the responses when I

02:47

tried cla a while back I cannot say that

02:50

I noticed that many incorrect responses

02:53

there were some but not very common I

02:55

think they mostly refer to Enterprise

02:58

cases where the type type of requests is

03:00

different and the amount of those

03:02

requests is much higher than anything a

03:04

regular person could ever reach Opus and

03:06

Sonet are already available to use as

03:09

API in 159 countries Hau will be

03:12

available soon but that isn't the only

03:14

interesting announcement that happened

03:16

stability AI has announced stable

03:18

diffusion 3 in early preview and the

03:20

newly published research paper promises

03:22

some seriously interesting stuff first

03:25

stable diffusion 3 uses

03:28

mmdi architecture which stands for

03:30

multimodal diffusion Transformer this

03:33

architecture is similar to the one used

03:35

by Sora previous versions of stable

03:37

diffusion and most current image

03:39

generation AIS use a diffusion model

03:42

large language models for text

03:43

generation like gbt use a Transformer

03:46

architecture diffusion models perform

03:48

well at creating detail in small regions

03:51

but are poor at generating the overall

03:54

layout of an image but Transformers are

03:56

good at layout but poor at Crea in

03:59

detail being able to combine the two

04:01

models lets us take advantage of the

04:03

best bars of both architectures and now

04:05

let's quickly look at example images

04:07

from the analment focusing on

04:09

limitations for example one of the

04:11

persistent issues for all AI image

04:13

generators is the text generation even

04:16

though stable diffusion 3 is much better

04:18

at this than it was before we can still

04:20

see imperfections that a person wouldn't

04:22

allow to happen these imperfections come

04:24

down to spacing of letters keeping them

04:27

properly sized and so on for for example

04:30

the main image with the text stable

04:32

diffusion 3 has a couple of these issues

04:35

like letters A and B becoming one letter

04:37

and variable distance between letters

04:39

and the image of a Bus shows the size

04:42

inconsistency on the dreamon sign all

04:45

the letters have a different size also

04:47

on the same image we can see that stable

04:50

diffusion still isn't perfect when

04:52

working with shadows the Shadow from the

04:54

bus Falls in One Direction while the

04:56

Shadow from the houses fall in a

04:58

different direction however the

05:00

developers in their research paper put a

05:02

huge emphasis on performance and human

05:04

feedback we have compared output images

05:06

from stable Fusion 3 with various other

05:09

open models as well as closed Source

05:11

systems such as Dolly 3 moury V6 and

05:14

idag V1 during these tests human

05:17

evaluators were provided with example

05:19

outputs from each model and as to select

05:22

the best results based on how closely

05:24

the model outputs follow the context of

05:26

the prompt it was given how well text

05:29

was rendered based on the prompt and

05:31

which image is of a higher aesthetic

05:34

quality we have found that stable

05:35

diffusion 3 is equal to where

05:37

outperforms current state of the art

05:39

text to image generation systems in all

05:42

of the above areas all of this sounds

05:44

really promising but unfortunately stab

05:47

Fusion 3 isn't publicly available yet

05:50

but as soon as it lands we will make a

05:52

video about it so you better you better

05:54

subscribe to the channel another

05:55

announcement that just happened was

05:57

co-pilot for finance from Microsoft this

06:00

is a very nice niche product optimized

06:02

for people working in finance with huge

06:04

spreadsheets from what I see on the

06:06

website it seems eily similar to those

06:09

wallframe logins for Chad gbt and it

06:12

will be integrated into the Excel

06:14

Outlook and teams and should help with

06:16

routine tasks and data analysis for

06:18

example it can access and analyze data

06:20

from various sources including Financial

06:22

systems like Dynamics 365 and sap as

06:26

well as emails and spreadsheets it also

06:28

can automate repetitive tasks such as

06:30

variance analysis Collections and

06:32

invoice matching it can generate

06:34

insights and recommendations and even

06:36

suggest actions Based on data this all

06:38

sounds really great especially for

06:40

people working the corporate jobs in the

06:42

financial area doing less manual work

06:44

being more efficient making more

06:46

informed decisions doesn't sound a

06:49

little dystopian to you we made a video

06:51

about jobs that may soon disappear and

06:53

seems like Finance professionals should

06:55

be added to that list now there's also

06:57

been a very interesting Chad gbt arrival

06:59

revealed mistol large was created by a

07:02

French startup mistro and Promises to be

07:05

very similar to gbt 4 according to the

07:08

company Mr Large ranks second in the top

07:10

llm Global list with metrics and

07:13

accuracy only outperformed by Chad bt4

07:16

the developers highlight a couple key

07:18

capabilities it's natively Flint in

07:20

English French Spanish German and

07:23

Italian with a Nuance understanding of

07:25

grammar and cultural context it has a

07:28

32k token context window precisely

07:31

follows the instructions and allows for

07:34

custom moderation policies and it's even

07:36

natively capable a function calling but

07:39

let's look at the comparison chart and

07:41

see whether it is as good as the

07:43

developers claim it to be and judging by

07:45

the numbers mistal is pretty close to

07:47

CAD bt4 in common sense reasoning and

07:50

knowledge and when it comes to

07:52

multilingual capacities mol outperforms

07:55

mexal and Lama 2 by a noticeable margin

07:58

even in terms of math and coding it is

08:01

similar to gbt 4 and Gemini Ultra the

08:04

developers are also releasing a small

08:06

and more optimized model focusing on

08:08

latency and cost and apparently this

08:10

model is also a tbit faster than its

08:13

competitors I think it is great that

08:14

European startups are catching up to the

08:17

big players in that industry we need

08:19

more news like that even though this

08:21

startup doesn't have that much money or

08:24

capacities as open AI or Google its

08:26

partnership with Microsoft is quite

08:28

promising and could lead to interesting

08:30

developments that are different from one

08:32

is doing it would be really fun to make

08:35

a huge comparison of the most powerful

08:37

AI Chad Bots don't you think the world

08:39

of music is also moving forward with

08:41

Adobe previewing Cutting Edge generative

08:44

AI tools for crafting and editing custom

08:47

audio the new project titled music gen

08:49

AI Control according to the post allows

08:52

musicians to generate music from text

08:54

Broms and even have full control over

08:56

the audio with Project Music gen AI

08:58

Control generative AI becomes your

09:01

co-creator it helps people crack music

09:03

for their projects whether they are

09:05

broadcasters or podcasters or anyone

09:08

else who needs audio that's just the

09:10

right mood tone and length the AI will

09:12

then create an initial tune based on the

09:15

users's prompt which they can also edit

09:17

using text Adobe set users can for

09:20

instance edit the generate music's

09:22

intensity extend the length of the music

09:24

clip or create a repeatable look I must

09:27

say the video they made doesn't really

09:29

show anything visually but it's fun to

09:32

watch also the tool doesn't even have

09:34

the interface yet it's all done through

09:37

the console commands this is very

09:39

different from commercially available AI

09:41

music generators we've tested before and

09:43

some of them even give good results for

09:46

example music effects created by Google

09:48

is a formidable opponent to adobe's

09:50

Creation so it might hint at what the

09:52

finished product will look

09:58

like

09:59

[Music]

10:06

music effect generates tracks from a

10:08

text prompt and lets users control each

10:11

part of the prompt through drop- down

10:13

menus I cannot say that this is some

10:15

crazy flexibility but it's already

10:18

enough to create some background music

10:19

for a short video we've also tested

10:21

beatoven and it was quite good this was

10:24

an interesting week for sure and one of

10:26

the releases was Amazon's rofus even

10:29

though the initial unveil happened in

10:31

February only now it has gotten to a

10:33

point where people were actually able to

10:35

test it and use it and well it's

10:38

definitely something here is what Amazon

10:40

says about it Rufus is an expert

10:42

shopping assistant trained on Amazon's

10:44

product catalog and information from

10:46

across the web to answer customer

10:47

questions and shopping needs products

10:49

and comparisons make recommendations

10:51

based on this context and facilitate

10:53

product Discovery in the same Amazon

10:55

shopping experience customers use

10:57

regularly but when the Washington Post

11:00

did their testing they found that Rufus

11:02

is not quite there to be called an

11:04

expert Amazon's chatbot doesn't deliver

11:07

on the promise of finding the best

11:09

products for your needs or getting you

11:10

started on a new hobby depending on how

11:13

I phrase the question the Amazon bot

11:16

several times offered basic suggestions

11:18

that I could find in a how to article

11:20

and didn't recommend specific products

11:22

still when the Amazon bot responded to

11:24

my questions I usually couldn't tell why

11:27

the suggested products were considered

11:28

the right ones for me or I didn't feel I

11:31

could trust the Chad bot's

11:32

recommendations to me this all is quite

11:34

fun huge corporation with all the money

11:37

in the world rolling out an

11:38

underdeveloped and untested feature just

11:41

for the sake of it we get to laugh and

11:43

the users get frustrated it's definitely

11:45

not something we want AI to do luckily

11:48

ruus is still under development so

11:50

hopefully in a few months it will

11:52

eventually become somewhat decent but

11:55

not every company is as bad at AI as

11:57

Amazon Wix a website building platform

12:00

has just released their own AI Builder

12:02

and it's pretty cool creating a site is

12:04

as easy as hitting the create with AI

12:07

button and chatting with the bot about

12:09

what you're looking for in your website

12:11

the questions are quite simple and take

12:13

only a couple of seconds to answer

12:15

basically it's like chatting with a

12:16

human designer then you wait a bit and

12:19

get a full-fledged website the promo

12:21

materials in my opinion are bit

12:23

exaggerating the real capabilities of

12:25

the AI in past videos I've tested dozens

12:28

of different AI website Builders and

12:30

none of them were able to generate

12:32

something like that usually the generate

12:34

websites are quite crude and simple in

12:37

their nature without complicated

12:38

animations or anything like that Wix

12:41

also can change themes for the website

12:44

Each of which has a different color

12:45

palette and overall look but if you want

12:48

you can just type in more requirements

12:49

for the website and will think of

12:51

something even the layout can be changed

12:54

with a simple prompt I like that very

12:56

much this promotional video really

12:58

paints a very very pretty picture but if

13:00

we take a look at the website we'll see

13:02

that the real world results are a bit

13:04

more modest still nice but modest don't

13:07

expect anything Super Creative or

13:09

Advanced this is pretty much the level

13:11

of a junior designer but faster and

13:13

cheaper and Wix is a very well-known and

13:16

reputable platform for website building

13:18

so I have a lot of trust in this Ai and

13:20

believe that it will get better and

13:22

better with every update there's also

13:24

been a small update to the GPT store the

13:26

new update allows users of 30 party gpts

13:29

to rate them using a drop- down menu

13:31

that lists a new Option send feedback

13:34

open AI showed how it's done and

13:36

consensus a pretty good research

13:38

assistant we should make a video about

13:40

it anyway to leave a review we need to

13:42

click on the GPT name and select send

13:45

feedback the menu will then appear

13:46

allowing us to leave a one to five star

13:49

rating and a text box to send an

13:51

optional private email to that gbts

13:54

Creator there's also a checkbox option

13:56

where we can leave our own address

13:58

visible or hidden a little privacy trick

14:00

I know this feature seems small and

14:02

insignificant but you won't believe how

14:04

many interest in gbts I've tested that

14:07

needed some fine-tuning and adjustment

14:09

and not all of them had an option to

14:11

contact their creator in any way now

14:13

when you encounter a custom gbt and have

14:16

problems with it or have ideas on how to

14:18

improve things you can send all your

14:19

suggestions directly to the person who

14:22

created this gbt also this rating will

14:24

probably be used in the store indicating

14:26

user satisfaction with certain gbts

14:29

again this is a small update but an

14:31

important one but open a felt quite

14:34

talkative this week and released a blog

14:36

post about its relationships with Elon

14:38

Musk and Tesla and you won't believe

14:40

what's inside I'm not going to read you

14:42

the whole thing but instead I'll will

14:43

give you a shorter version what happened

14:45

between Elon Musk and open Ai and how

14:48

Tesla was going to acquire open AI

14:50

please know these events transpired a

14:52

few years back and were published only

14:54

now following elon's lawsuit so here's

14:58

the story open AI is all about making

15:00

sure artificial general intelligence AGI

15:03

is good for everyone they realized early

15:05

on that building AGI would be a massive

15:08

money sync far beyond initial estimates

15:11

Elon was one of the big names behind

15:13

their funding suggesting they announc a

15:16

whopping $1 billion commitment to kick

15:19

things off his pitched in less than 45

15:22

million himself with over 90 million

15:25

coming from other supporters at first

15:27

the team thought $100 million might be

15:31

enough but Elon pushed for a bigger

15:33

figure to not look weak next to Tech

15:35

giants like Google by 2017 it became

15:38