-2.svg){kind=logo}
The past month has witnessed rapid developments in generative AI that marketers must stay on top of to remain competitive. This roundup summarises crucial takeaways from episode 31, episode 32, episode 33 and episode 34 of the acclaimed "Artificially Intelligent Marketing" podcast, hosted by preeminent experts Paul Avery (our CEO) and Martin Broadhurst.
Spanning trailblazing model upgrades like OpenAI’s overhauled ChatGPT and DALL-E editions to Google’s new conversational AI capabilities and Adobe’s enhanced creative tools, opportunities to leverage AI continue rapidly expanding across content creation, data analysis, team collaboration and more. However, the hosts also responsibly highlight important considerations around issues like responsible AI adoption, data quality, human oversight of outputs, and emerging legal grey areas that require further guidance. With new models constantly launching amid enthusiastic uptake, taking time to properly evaluate and intimately understand these technologies remains crucial.
This blog contains high-level overviews summarising the main discussions across the November podcast episodes. For full details and context, we encourage you to listen to episodes 31, 32, 33 and 34, available below or on your preferred podcast platform.
Summaries from episode 31:
HubSpot Acquires Clearbit for Enhanced B2B Data
HubSpot has acquired Clearbit, a leader in B2B data and intelligence. Their database covers millions of companies with firmographic, demographic, and technographic insights. The acquisition will bake this data directly into HubSpot's CRM platform, unlike any other on the market. This could enable AI-powered functionality like customer segmentation and personalised messaging.
OpenAI Upgrades ChatGPT Interface
OpenAI is upgrading the ChatGPT interface so users can access capabilities like image generation, plugins, and data analysis without switching modes. This allows seamless combinations like using an image to inspire another image. The changes may make many third-party ChatGPT plugins redundant, creating tension between OpenAI improving its product and relying on external developers.
Stability AI Launches New Image Capabilities
Stability AI is introducing products for industries like real estate and gaming. One feature automatically replaces skies in photos to always show bright blue skies. Another enables rapid 3D model generation to reshape architectural visualisation and video game asset creation. They also announced Stable Fine-Tuning to customise aesthetic outputs.
Emotional Prompts Enhance AI Performance
New research shows emotional prompts like urgency and motivation improved accuracy 8-115% over baseline models on ML benchmarks. Human raters also preferred outputs from emotional prompts, saying a poem exhibited more creativity and narrative flow. This suggests basic emotional intelligence in large language models.
Woodpecker Framework Tackles AI Hallucinations
The Woodpecker framework aims to reduce false information from large language models using multi-step validation against visual data. As hallucinations undermine trust in AI, approaches leveraging external knowledge to correct inaccuracies hold promise for improving reliability.
AI Detects Diabetes from Voice
Canadian researchers developed an AI model diagnosing Type 2 diabetes from voice recordings alone. By detecting subtle acoustic patterns imperceptible to humans, it demonstrates AI’s unique ability to uncover disease biomarkers in unexpected data signals like audio.
AlphaFold 2 Predicts Biological Structures
DeepMind launched AlphaFold 2, improving predictive accuracy for protein structures while expanding model capabilities to additional molecules like DNA and RNA. Further advancing computational biology, it brings fuller realisation of initiatives like digital drug discovery closer to reality.
Bletchley AI Summit Focuses on Risks
At the Bletchley AI summit, global experts signed an agreement governing ethical development while discussing existential risks. The event included participation from open source providers to balance perspectives from big tech companies arguing for regulation.
Microsoft Rolls Out Windows 11 Co-Pilot
Windows 11 Co-Pilot allows users to accomplish tasks conversationally. For example, it can enable do-not-disturb mode through voice commands rather than digging into settings menus. This aims to help non-power users better leverage Windows capabilities.
Office 365 Co-Pilot Available to Enterprises
Microsoft introduced an AI co-pilot for Office 365, initially launching through enterprise licenses costing $30 per user monthly. Early access suggests it currently has limitations versus expectations, struggling with nuanced tasks. But steady improvements over time could make it indispensable.
Summaries from episode 32:
Episode 32 was a bonus episode covering key updates from the OpenAI Developer Conference on 6th November 2023.
GPT-4 Turbo Launched with 128k Context Window
OpenAI launched GPT-4 Turbo with a drastically upgraded 128,000 token context window, enabling feeding entire 300-page documents or lengthy PDFs to deeply analyse content - previously impossible within chatGPT's 4,000 token limit. However, testing indicates memory and recall declines around 70,000 tokens, struggling for passage middles. The upgraded capacity reaches ChatGPT Plus users and is offered at ~2.7x lower computational pricing to developers, though high context usage from 128k windows can still quickly drive up costs by $1 per query.
New Assistant API Enables Specialised AI Agents
OpenAI introduced an Assistant API allowing developers to build multi-modal AI apps leveraging models like DALL-E for images, GPT-4 for text, the new Whisper v3 for speech recognition, and more. This opens creative possibilities like specialised coding assistants, AI-powered vacation planners, voice-controlled DJ programs, smart canvas apps for any idea. A new retrieval system also reduces repetitive context re-explanations by remembering user preferences across conversations rather than just prompt inputs.
No Need to Switch Modalities in ChatGPT Plus
ChatGPT Plus now automatically handles images, data analysis, and other tasks without needing to toggle between modes by selecting specialised tools from a dropdown menu. However, it sometimes incorrectly activates modalities like DALL-E image creation when not intended, introducing friction into workflows rather than smoothing operations. OpenAI acknowledges this version is not yet solving for specific user problems, instead providing building blocks for users to create solutions.
Free ChatGPT Users Get GPT-3.5 Turbo
A new GPT-3.5 Turbo version with 16,000 token context window will be provided for free chatGPT users, combined with the 128k window for paying Plus and Enterprise tiers. This continues OpenAI's computational optimisation efforts to deliver increasingly advanced applications at lower costs, while stratifying platform access tiers primarily by context length.
Retrieval Tool Enables User Memory
A new retrieval tool for apps built on OpenAI allows remembering user context, preferences and state across conversations instead of just prompt inputs. This reduces repetitive explanations of user history, aiding workflow. As tools leverage these updates, expect improved memory and relevance aligned to historical dialogues rather than solely immediate prompt content.
OpenAI Launches GPT Store for Sharing Bots
OpenAI introduced a GPT store where anyone can build specialised assistants and share custom chatbots without coding using an intuitive wizard interface. Instruction priming, knowledge infusion from documents, and integrated tool activation enables niche personalisation. Popular community-created GPTs will rank on a leaderboard, incentivising distribution with usage-based monetisation. This democratises access for non-developers to leverage OpenAI for targeted use cases.
Copyright Shield Protects Customers
OpenAI launched Copyright Shield, agreeing to cover any legal costs if customers face copyright lawsuits around generally approved platform uses. This protection helps assure enterprises can confidently deploy AI capabilities without risks of infringement claims, echoing similar legal indemnification from Microsoft and Google around their language models.
Whisper Speech Recognition Updated to v3
OpenAI announced Whisper version 3, updating their foundational automatic speech recognition model that powers existing interactive capabilities like chatGPT mobile app conversations. While the current edition is already impressively capable, further improvements to Whisper's speed and accuracy could enhance these voice-driven workflows.
Summaries from episode 33:
ChatGPT Updates - A Deep Dive into the Latest Features
ChatGPT has rolled out a 128,000 token context window, reduced API pricing, and consolidated modalities into one interface. However, real-world testing shows recall declines around 70,000 tokens, and frustrations persist like inconsistent tool selection and throttled capabilities. Performance issues also plague the platform amid surging demand, challenging reliability.
The Rising Star of ChatGPT
ChatGPT's meteoric rise has led to infrastructure strains impacting stability and functionality. Problems include downtime, slow outputs, reduced image generation, lower message limits, and confusing multimodal behaviours. OpenAI acknowledges not solving for specific user problems, instead providing building blocks for users to create solutions.
Microsoft Ignite 2023: The Dawn of Copilot Studio
Microsoft announced Copilot Studio, allowing no-code customisation of Copilot AI assistants across enterprises. This hints at an emerging role of AI architect to design systems leveraging models like GPT-4. New computer vision and voice avatars also showcase multimedia potential, though realism makes differentiating real content challenging.
Unpacking GPT-4's Memory Capabilities
Tests show GPT-4 Turbo's 128,000 token context window degrades after 70,000 tokens, struggling with recall from passage middles. So the upper limit far exceeds chatGPT's prior 4,000 tokens, but falls short of claimed capacity for comprehending book-length inputs. Still, the boost benefits many use cases.
Google's Anticipated 'Gemini' AI: A New Challenger Awaits
Google postpones its next-gen “Gemini” conversational AI model launch to further refine capabilities before challenging chatGPT supremacy. With few details known, potential impact across search, recommendation and broader assistance remains ambiguous for now.
Google's Global AI Search Expansion
Google expands its AI-enhanced Search Generative Experience globally. This could profoundly reshape search behaviours as the system directly answers queries rather than linking relevant sites. Expect content creators to adapt for prominence as AI penetrates the discovery process.
YouTube's Commitment to Responsible AI
YouTube plans increased regulation around synthetic media, including mandatory disclosures of AI use. Labels, policies, and tests will aim to curb misuse and misinformation while fairly compensating creators. This will likely impact marketers using generative voice, image, and video tools.
YouTube's Melodic AI: Tuning into the Future
YouTube and DeepMind launched Melodic AI to enable track creation by humming melodies and describing desired instruments, moods, and more. This provides new creative tools for musicians and video producers. Meanwhile, an ethics resignation at Stability AI highlights tensions around data usage.
Ethics in AI Training: A Resignation at Stability AI
Ed Newton-Rex, Head of Audio at Stability AI, resigned over what he deemed exploitative and exclusionary AI training practices using copyrighted data lacking artist permission or compensation. His exit spotlights building tensions around generative model development methods involving scraped materials.
Runway's Motion Brush: The Next Leap in AI Video Generation
Runway unveiled Motion Brush allowing users to designate moving elements within static images to create dynamic scenes. If effective, it could transform applications from lifeless charts to lively family photos. This increased access to video creation hints at more animation in marketing visuals.
OpenAI's Collaborative Vision with Data Partnerships
OpenAI invited proprietary data contributions like specialised museum collections to improve models like GPT-5, provided strict usage contracts govern access. Recognition technologies can ingest diverse inputs. This two-way value exchange gives institutions distribution while enhancing AI capabilities.
Summaries from episode 34:
Episode 34 was a special edition episode, where the hosts delve into AI's transformative role in marketing and beyond. Join Martin as he uncovers the impressive strides made by Eric Xiao and his AI tech company, Laxis.
Venture to Innovation: Discover Eric's Transition from VC to Founding Laxis
As a venture capitalist, Eric Xiao spent excessive time researching startups and founders before meetings, intensely focused on conversations without notetaking, then manually writing lengthy due diligence reports. Upon realising revenue teams like sales, business development and consultants wasted hours across similar pre-meeting, in-meeting and post-meeting work, Eric left VC to launch Laxis in 2020 to automate these workflows. By automatically capturing meeting notes and generating follow-ups, Laxis saves users 5+ hours per week.
Intelligent Integrations: Laxis's Impressive Capabilities
Laxis offers AI-powered meeting transcription so users never miss critical details and can focus fully on customer conversations without notetaking worries. Laxis’s AI writer additionally drafts relevant follow-ups like requirement documents, status updates and personalised outreach emails. Integrations with CRMs like Salesforce and HubSpot also automatically log notes, tasks, and details into contact records based on meeting outcomes. And a new Laxis Team Channel lets managers review sales call transcripts and provide feedback.
Laxis at a Glance: Boasting 25,000 Active Users
After just 2 years in market, Laxis already serves over 25,000 active individual users across over 3,000 paying organisations, primarily targeting small revenue teams of 5-20 people. While intended for sales and business development, Laxis discovered additional demand from marketing, product and other roles seeking access to meeting transcripts and insights to inform messaging and roadmaps. With compelling visibility into customer conversations, AI assistants like Laxis are uniquely enhancing efficiency and relationship-building.
Security in AI: Strict Protocols Protect User Data
Laxis addresses privacy concerns through encrypted storage of all customer conversation data on secure, industry-standard cloud platforms like Google and Microsoft. Though skepticism remains in some regions like Europe, recorded calls are increasingly common across contexts from sales conversations to customer service. As AI assistance permeates workflows via bots and wearables, Laxis maintains leading data governance best practices to earn user trust.
Marketing Mastery: Harnessing Organic Growth and AI Content
Laxis fueled early growth through organic marketing, including Google/Cisco co-marketing campaigns, AppSumo partnerships, affiliate programs with influencers, and relationship-based incentives for referrals and AI-powered meeting summaries. With early AI adopters as customers, Laxis also leverages its conversational transcription capabilities for podcast and content marketing to boost brand visibility, using AI to optimise blogs and SEO.
The Market of Minds: AI Assistance Adoption Still Early Stage
The increasingly crowded AI assistant space demands sharp vertical focus and broad platform availability. As a cross-platform startup, Laxis zeroes in on revenue teams spanning meeting systems, with plans to expand features and CRM integrations. With big players like Microsoft and Zoom launching their own branded Copilots, startups need specialised differentiation. Though adoption is still early stage, AI permeation will inevitably accelerate as capabilities improve.
Explore AI's Impact on Marketing with the "Artificially Intelligent Marketing" Podcast
This blog offers a glimpse into the insightful discussions hosted by Paul and Martin in their captivating podcast series. For a deep dive into how AI influences marketing, we encourage you to tune into the full episodes, available both below and on popular podcast platforms. Additionally, exploring the archive of past episodes will provide a wealth of knowledge on AI's transformative effect on marketing.
AI Note
This blog post was crafted with the aid of AI technology. Initially, AI was used to transcribe the podcast audio. Following this, Claude from Anthropic and GPT-4 (ChatGPT) assisted in refining the transcript into clear, concise blog content. The BioStrata team then meticulously edited these segments to ensure greater clarity and readability. Additionally, Clipdrop AI software played a key role in creating the blog's standout featured image.
