AI生成工具的新衝擊 - MS Bing & Google Bard 能否挑戰ChatGPT-4領導地位

AI - text-to-text /
text-to-image /
text-to-speech /
text-to-video / SEO
Ivan so

Ivan so
● 18 years of SEO and WordPress experience
● Built 50 site to test SEO
● Handled over 400 WordPress design and
development
● Lead organiser of WordCamp, co-organiser
of HK WP meetup and HK elementor
meetup

My bio
● 5 times amazon ebook best sellers
● Printed book in HK
● Mailchimp HK partner
● Mailerlite HK partner
● Shopify Partner

Client for consulting / training service.

index
1. What is AI
2. Text-to-text
3. Text-to-image
4. Text-to-speech
5. Text-to-video
6. Q&A

Artiﬁcial intelligence (AI) is intelligence—perceiving, synthesizing, and inferring
information—demonstrated by machines, as opposed to intelligence displayed by
non-human animals and humans. Example tasks in which this is done include speech
recognition, computer vision, translation between (natural) languages, as well as
other mappings of inputs.
AI applications include advanced web search engines (e.g., Google Search),
recommendation systems (used by YouTube, Amazon and Netﬂix), understanding
human speech (such as Siri and Alexa), self-driving cars (e.g., Waymo), generative or
creative tools (ChatGPT and AI art), automated decision-making and competing at
the highest level in strategic game systems (such as chess and Go).

Spot object - https://segment-anything.com/demo

AI生成工具的新衝擊 - MS Bing & Google Bard 能否挑戰ChatGPT-4領導地位

Tesla - data collection (bad data)

Chatgpt compete with programmers with result level, not on the code level.
Also, tech companies are not competing with quality of code but user attention span

Generative AI refers to a set of deep-learning technologies that use existing content, such as
text, audio, or images, to create new plausible content that previously would have relied on
humans. Generative AI is driven by unsupervised and semi-supervised machine learning
algorithms capable of identifying underlying patterns present in the input to generate similar
content, delivering innovative results without human thought processes or bias.
LLM (large language model) & machine learning
Source: Cassie Kozyrkov

Well-known examples of generative AI models include Open AI's GPT-3 (text generator)
and DALL-E (text-to-image generator) and Google's BERT (language model).
Various techniques are utilized by generative AI algorithms, and the two most widely used
models are generative adversarial networks (GANs) and transformer-based models.
Another popular technique is variational auto-encoders (VAEs).
AI models

ChatGPT is a word prediction model

How are _______________
It’s in our language after study

ChatGPT is a word prediction process

LM Neural network LM
Traditional language models are often
based on probabilistic models such as
n-grams, which rely on counting the
frequency of words or sequences of words
in a training corpus
Neural network language models use
neural networks to learn the patterns and
relationships between words
automatically from the data.
LM (language model) & Neural network LM
Neural network LM more complex patterns and dependencies between words in a sequence.

NLP
NLP (Natural Language Processing) is a subﬁeld of AI that
deals with the interaction between computers and
humans using natural language. The goal of NLP is to
enable computers to understand, interpret, and generate
human language.
Before the advent of NLP, computers could only process
structured data such as numbers and symbols. However,
with NLP, computers can now process unstructured data
such as text, speech, and images. This has enabled a range
of applications such as chatbots, language translation,
sentiment analysis, and text summarization.

Some examples of NLP
applications:

Sentiment Analysis
Sentiment Analysis: NLP is used to analyze the sentiment
of text data, such as social media posts, customer reviews,
and news articles. Sentiment analysis is used to
understand the opinion or emotions expressed in the text,
which can be useful for businesses to improve their
products and services. For example, a company may use
sentiment analysis to understand customer feedback on
social media and improve their product accordingly.

Chatbots
Chatbots: NLP is used to build chatbots that can
interact with humans using natural language. Chatbots
are used in customer service, sales, and support, and
can handle simple queries and tasks. For example, a
chatbot for a bank can help customers with their
account balances, transaction history, and transfer
funds.

Speech Recognition
Speech Recognition: NLP is used to recognize and
transcribe human speech into text. Speech
recognition is used in personal assistants like Siri and
Alexa, voice-controlled devices like smart speakers,
and in call centers to transcribe customer calls. For
example, a user can ask Siri to set an alarm, play music,
or read the news.

Named Entity Recognition
Named Entity Recognition: NLP is used to identify
and extract named entities, such as names,
organizations, locations, and dates from text.
Named entity recognition is used in information
retrieval, text mining, and information extraction.
For example, a news aggregator can use named
entity recognition to identify the names of people,
companies, and locations in news articles.

Text Summarization
Text Summarization: NLP is used to automatically
summarize long documents into shorter
summaries. Text summarization is used in news
articles, research papers, and legal documents. For
example, a news aggregator can use text
summarization to generate a summary of news
articles for readers.

Ai writing tool (ChatGPT, POE app,
Bing edge browser, Bard, 文心一言)

Chatgpt (vpn + overseas number) - https://chat.openai.com/chat

Chatgpt limitation. No internet / data before sep 2021 / moral / memory

Long short-term memory (LSTM) is an artificial neural network used in the fields of artificial intelligence and
deep learning. Unlike standard feedforward neural networks, LSTM has feedback connections. Such a recurrent
neural network (RNN) can process not only single data points (such as images), but also entire sequences of data
(such as speech or video).
This characteristic makes LSTM networks ideal for processing and predicting data. For example, LSTM is
applicable to tasks such as unsegmented, connected handwriting recognition, speech recognition, machine
translation, speech activity detection, robot control, video games, and healthcare.
LSTM - long short-term memory

Chinese content
Sometimes stop as not long enough
=> Continue
Continue if not complete
E.g. write a 1000 word about how to become a great basketball player. output language: traditional chinese

Add keyword 生蛇階段 in content

Add <h2> and <h3> from below content

叩門信
用以下資料為我的6歲兒子寫一份小一叩門信
中文姓名：陳小明
英文姓名：Chan Siu Ming, Joe
幼稚園：香江幼稚園（ K1-K3）
課外話動：畫畫、彈琴
喜歡運動

CV 應徵職位：明報記者
應徵要求：
1. 持有大學學位，中、英文良好，富責任心
2. 對本港、中國及國際時事及生活文化有濃厚興趣
3. 喜好夜班工作，喜好白紙黑字的閱讀質感
中文姓名：陳大明
英文姓名：Chan Tai Ming, Peter
電郵：xxxxxxxxxx@123.com
地址：九龍黃大仙 XX樓XX室
工作經驗：
5/2019-1/2020 香江電台製作助理
- 負責節目前期資料蒐集
- 接待主持人及嘉賓
- 接剪錄音

https://poe.com/

Exercise - Research
assistance
Task: Participants research different
renewable energy sources (solar, wind,
hydro, geothermal, etc.) and their
environmental impacts.
Exercise: Use ChatGPT to gather
information on each energy source,
including advantages, disadvantages, and
recent developments.

Exercise - Drafting
and editing
Task: Participants write a short news article
about a new renewable energy project in a
speciﬁc location.
Exercise: Use ChatGPT to help draft the
article, generate headlines, and suggest edits
for improved clarity and style.

Exercise - Story
ideas and
brainstorming
Task: Participants brainstorm story ideas
related to climate change and renewable
energy.
Exercise: Use ChatGPT to generate a list of
potential story ideas, angles, or interview
subjects related to the given topic.

Exercise - Interview
preparation
Task: Participants prepare to interview a
renewable energy expert.
Exercise: Use ChatGPT to gather
background information on the expert, their
ﬁeld, and formulate relevant questions to
ask during the interview.

Exercise - Social
media and audience
engagement
Task: Participants create social media posts
to promote their renewable energy articles.
Exercise: Use ChatGPT to generate engaging
captions, hashtags, and headlines for various
social media platforms.

Exercise -
Language
translation
Task: Participants translate a paragraph
from a foreign language article on renewable
energy into English.
Exercise: Use ChatGPT to translate the text
and compare the results with the original
content to assess translation accuracy.

Exercise - SEO
content format
Add <h2> and <h3> for following content
xxxxx

Exercise - keyword
research
List me suggested keyword, related keyword
and keyword cluster for ‘keyword’

Exercise
(fast learner mode)
Find out what is GAN

Fast learner mode for research. E.g What is GAN

Some common useful prompt engineering
techniques are:
1. Providing context & task specifications (Instruct prompting). Include context of what
you’re trying to do and ask specifically about the type of content and format you want to get
from the language model.
2. Few-shot prompts. Showing the language model a couple of prompt-answer pairs as
examples in your prompt can greatly improve the quality of responses. Include diverse and
representative examples similar to the actual query you’re asking.
3. Chain-of-Thought (CoT) prompting. This technique is helpful for complex reasoning tasks. It
asks the LLM to generate a sequence of logical reasoning steps and reach the final answer .
Researchers found attaching “Let’s think step by step” at the end of the prompt works well.
4. Generated Knowledge/Internal Retrieval. This technique asks the LLM to generate knowledge
that could potentially solve the task before solving the task. For example, attaching “Generate
some knowledge about the query to answer the question” before your actual prompt can
improve common-sense reasoning in LLMs.

Advanced methodology of
prompt

Learn from template - https://cohesive.so/

resources
https://medium.com/mlearning-ai/the-chatgpt-list-of-lists-a-collection-of-1500-useful-mind-blowing-an
d-strange-use-cases-8b14c35eb

https://gist.github.com/coolaj86/6f4f7b30129b0251f61fa7baaa881516

Chatgpt extension -
https://chrome.google.com/webstore/detail/aiprm-for-chatgpt/ojnbohmppadfgpejeebfnmnknjdlckgj

Outrank article. Test - https://www.hdcourse.com/testing-page/
Prompt: outrank the competition with an in-depth, seo-optimized article based on 'I go to
school by bus'. be like our competition, just a little better

Poe app by quora (ios) use openai & anthropic - https://poe.com/

Web - https://poe.com/

Create a bot
Check this out on Poe: https://poe.com/emojiai

Earn poe backlink for seo - https://poe.com/s/pnXmGgdfbLKqxtMyqVFi

https://www.anthropic.com/

Prometheus (better than ChatGPT)
Basically, the technology is an AI model that Microsoft created to combine the Bing index,
ranking, and answers search results with OpenAI’s GPT models. This makes the ChatGPT
models have fresher, almost real-time, content and data to use for its training models.
It also can read PDF

Prometheus story
Prometheus is best known for defying the Olympian gods
by stealing ﬁre from them and giving it to humanity in the
form of technology, knowledge, and more generally,
civilization. In some versions of the myth, he is also
credited with the creation of humanity from clay.

Bing edge + chat (beyond 2021) - https://www.bing.com/

cannot show images
Can show ads

Mar 2023. Openai trafﬁc surpass bing

Google bard thinks he/she is better than chatgpt. Need to join waiting list (less than a week waiting) and Uk/usa ip

文心一言 - generated content is very similar to existing content of other sites

Ecom: AI avatar generation with VO

3b / 7b: https://github.com/Stability-AI/StableLM

See the area of improvement to
see what you could do in those
area

2 generation is 100 pages which you can
publish a book easily

Summary - 98 pages technical report https://arxiv.org/pdf/2303.08774.pdf
1. Bing chat is using gpt-4 (daily 15 quota)
2. Content length x2. Some user x8 (50 pages of text)
3. Training model, dataset is hidden to avoid competition
4. Better exam result. ESP bar exam top 10% result. Gpt-3.5 was bottom 10%
5. 100% hindsight neglect 事後孔明 (un-expected) - prediction. Gpt-3.5 is under 20%
6. 30% prefer gpt 3.5 output (thousands of prompt test)
7. Improvement for languages
8. Multi modal (text / image input) - understand Infograph, chart (chartQA test)
9. Less hallucination (factual accuracy)
10. Data by sep 2021
11. Seeking power (machine learning not fully under human control) - By Arc research center. Code itself for
improvement

Gpt-4 has eyes (visual ability)

https://minigpt-4.github.io/ - supported multi-modal: pretrained MiniGPT-4 aligned with Vicuna-7B

Become a platform like wechat super app - https://openai.com/blog/chatgpt-plugins

Browsing via bing api. Threat to google

ChatGPT plugin development opportunities - https://platform.openai.com/docs/plugins/introduction (100M user)

Greg Brockman, co-founder of openai demo various plugins usage (just like prompt sequence)
● Case: write recipes, buy food (instacart), generate img (dalle-2), post to twitter (zapier)
● Case: user feedback help to improve the output quality
● Case: code interpreter - upload csv to analyse and then make graph with data

Stage X A Conversation with Bill Gates and Jessie Woolley-Wilson | ASU+GSV 2023 (Summit connects
leading minds around digital learning and workforce skills to ensure a new era) - Apr 2023
● To enhance children's reading ability and improve writing skills within 18 months
● Current AI chatbots can already read and write very ﬂuently
● Become as good as a private tutor as any human ever could
● All children can receive high-quality education regardless of their economic background
● Most students cannot afford one-on-one private tutoring, AI can help

Prompt I used
juicy Big Mac burger, ﬂying exploded view ,super slow
motion, award winning photography, food
advertisement, prime time ad --v 5

https://www.canva.com/ (template + text-to-image)

Exercise Design a poster
Monkey on the moon
Moon travel

https://www.canva.com/magic-write/

Exercise Write about eco friendly topic and convert
into PPT

https://www.canva.com/magic-design/

Exercise Insert media for different suggested layout
by Canva
Re-write text in video
Change image in ﬂash card

What is it?
An AI chatbot is a computer program that
uses artiﬁcial intelligence techniques, such
as machine learning and natural language
processing, to simulate human
conversation. It is designed to interact with
humans through text or voice-based
interfaces, such as messaging apps,
websites, and voice assistants.

What is it?
AI chatbots use algorithms to analyze and
understand the user's input, and then
generate appropriate responses based on
pre-deﬁned rules or learned patterns.
They can provide information, answer
questions, perform tasks, and even make
recommendations based on the user's
preferences and behavior.

What is it?
AI chatbots have become increasingly popular in recent years, as they
provide businesses and organizations with an efﬁcient and cost-effective
way to engage with customers and users. They can handle large volumes of
inquiries and provide 24/7 availability, improving customer satisfaction
and reducing the workload on human customer service representatives.

Contact
ivan@hdcourse.com
www.fb.com/ivansopage
www.youtube.com/c/ivanso

AI生成工具的新衝擊 - MS Bing & Google Bard 能否挑戰ChatGPT-4領導地位

Recommended

More Related Content

What's hot (20)

Similar to AI生成工具的新衝擊 - MS Bing & Google Bard 能否挑戰ChatGPT-4領導地位 (20)

More from eLearning Consortium 電子學習聯盟 (20)

Recently uploaded (20)

AI生成工具的新衝擊 - MS Bing & Google Bard 能否挑戰ChatGPT-4領導地位