Kozminski Techblog

Reading Time: 5 minutes

Large language models (LLMs) have emerged as a pivotal technology in artificial intelligence, spearheaded by innovations from Silicon Valley companies like OpenAI and Anthropic. Models such as GPT-3, Claude, and ChatGPT have demonstrated impressive natural language generation capabilities.

In response, European startups have begun developing alternative LLMs, with French company Mistral AI gaining traction as a leading contender. This paper compares technical specifications, funding, commercial partnerships, and market positioning of ChatGPT versus Mistral to analyze the latter’s viability as a European champion in generative AI against established US players.

• Model Architectures: ChatGPT utilizes a standard transformer-based neural network architecture with attention layers to process textual context. In contrast, Mistral employs a mixture-of-experts model that combines specialized sub-modules to enhance efficiency and performance. Specifically, Mistral’s Mixtral 8x7B model uses 46.7 billion parameters sparsely via router networks, resulting in equal accuracy as a dense 129 billion parameter model. This novel sparse training approach could yield sustainability advantages regarding environmental impact. Meanwhile, Claude boasts unique capsules and mixture objectives for robustness. On balance, Mistral appears favorable over ChatGPT regarding technical innovation but trails Claude currently.

• Benchmark Performance: Available benchmarks reveal Mixtral 8x7B matching GPT-3.5 and exceeding other models like LLaMA-2 70B across metrics including quality-latency tradeoffs and multi-task accuracy. Its instruction-following fine-tuning establishes state-of-the-art standards on established tests like TruthfulQA and MT-Bench, even besting Claude. Thus, Mistral’s implementations seem extremely competitive regarding critical functionality.

• Fundraising and Valuation: By virtue of OpenAI’s early dominance since its 2015 inception, Microsoft and other backers have invested over $10 billion establishing commanding market position. Comparatively, nine-month-old Mistral AI raised $500 million to date at a $2 billion valuation, reflecting bullish sentiment about its commercial prospects from top Silicon Valley investors like General Catalyst. The sheer velocity of capital accumulation highlights Mistral’s excellence and the vast market potential of generative AI. ChatGPT clearly retains resource advantages from extensive funding history, but Mistral’s extraordinary seed traction shows promise.

• Partnerships and Market Traction: ChatGPT originally leveraged Microsoft infrastructure but now diversifies across Google Cloud, Oracle, and AWS as adoption accelerates. Mistral AI has navigated partnerships with French cloud startups, Google Cloud, and IBM. Having announced collaborations with over 10 major companies spanning industries and geographies, Mistral is establishing commercial validity amidst competitive posturing between tech titans hoping to dominate associated cloud value chains.

The strength of Mistral AI partnerships and market tractionn of his has been particularly showcased by the recent investment and partnership of Nvidia with Mistral AI, which could be the key factor in allowing Mistral to successfully compete with the Silicone Valley LLMs. This is due:

1.     Hardware Advantage: Nvidia dominates the GPU market, the preferred chip for training and running large language models. This gives Mistral potential privileged access to the most advanced hardware for developing and deploying its models. Having the best infrastructure boosts iteration speed and model scales.

2.     Technical Collaboration: The partnership likely involves Nvidia assisting Mistral in optimizing its software and models to run maximally efficiently on Nvidia GPUs. This engineering collaboration could help Mistral match or exceed the performance of OpenAI’s models running on similar hardware.

3.     Cloud Distribution: Nvidia also provides cloud infrastructure and services to host AI models. Prioritizing and integrating Mistral’s offerings on Nvidia’s platform improves accessibility and reduces go-to-market friction against OpenAI’s presence on other clouds.

4.     Signal of Momentum: A stamp of credibility from the AI chip leader Nvidia further cements Mistral as a rising force, especially against OpenAI’s reliance on Azure, attracting more talent and investments into its ecosystem.

If executed thoroughly, this boost can be enough to make Europe home to the world’s leading generative AI powerhouse for the coming era.

Yet, despite early traction, Mistral AI lacks the infrastructure, resources, and market momentum to fully challenge the leadership of OpenAI, Anthropic, and other US-based LLMs. 

Arguments:

1.     Computational constraints: Generating cutting-edge LLMs requires access to immense computing power for model training that only global hyper-scalers like Microsoft and Google Cloud realistically provide. Mistral relies on 3rd party cloud services, limiting control over critical infrastructure.

2.     Commercial ecosystem: The sheer breadth and maturity of enterprise integrations, partnerships, distribution channels and developer communities underpinning the adoption of LLMs like GPT-3 and Claude take years to cultivate organically. Mistral’s commercial presence remains relatively narrow despite corporate POCs.

3.     Geopolitical fragmentation: Contending with disparate national-level AI policies and priorities across EU member states dilutes legislative support and resources that Mistral needs to maximize continental success before tackling global expansion.

4.     Talent consolidation: Silicon Valley’s concentration of expertise in training techniques, software frameworks and model architectures has compounded over nearly a decade into nearly insurmountable competitive advantage. Mistral must battle extreme talent scarcity.

5.     Computational constraints: Generating cutting-edge LLMs requires access to immense computing power for model training that only global hyper-scalers like Microsoft and Google Cloud realistically provide. Mistral relies on 3rd party cloud services, limiting control over critical infrastructure.

6.     Commercial ecosystem: The sheer breadth and maturity of enterprise integrations, partnerships, distribution channels and developer communities underpinning the adoption of LLMs like GPT-3 and Claude take years to cultivate organically. Mistral’s commercial presence remains relatively narrow despite corporate POCs.

7.     Geopolitical fragmentation: Contending with disparate national-level AI policies and priorities across EU member states dilutes legislative support and resources that Mistral needs to maximize continental success before tackling global expansion.

8.     Talent consolidation: Silicon Valley’s concentration of expertise in training techniques, software frameworks and model architectures has compounded over nearly a decade into nearly insurmountable competitive advantage. Mistral must battle extreme talent scarcity.

9.     Computational constraints: Generating cutting-edge LLMs requires access to immense computing power for model training that only global hyper-scalers like Microsoft and Google Cloud realistically provide. Mistral relies on 3rd party cloud services, limiting control over critical infrastructure.

10.   Commercial ecosystem: The sheer breadth and maturity of enterprise integrations, partnerships, distribution channels and developer communities underpinning the adoption of LLMs like GPT-3 and Claude take years to cultivate organically. Mistral’s commercial presence remains relatively narrow despite corporate POCs.

11.   Geopolitical fragmentation: Contending with disparate national-level AI policies and priorities across EU member states dilutes legislative support and resources that Mistral needs to maximize continental success before tackling global expansion.

12.   Talent consolidation: Silicon Valley’s concentration of expertise in training techniques, software frameworks and model architectures has compounded over nearly a decade into nearly insurmountable competitive advantage. Mistral must battle extreme talent scarcity.

Mistral AI’s combination of technical creativity, commercial validation, and geopolitical tailwinds position it strongly to emerge as a viable European alternative to established US generative AI ecosystems. Mistral AI has exemplified European ambitions in seeding a generative AI challenger but outflanking the Silicon Valley ecosystem likely requires order-of-magnitude, patient capital investment over years measuring progress in small increments rather than months. Consequently, near-term hype exceeding realistic capabilities risks disillusionment slowing broad LLM democratization. Sustained commitment to balance commercial viability and public interest is key. Yet, with rapid traction across benchmarks, fundraising, corporate customers, and global cloud leaders demonstrates execution excellence despite the profound resource asymmetry against OpenAI and Anthropic in the race to lead LLMs powering the AI economy. Sustaining momentum depends significantly on continued innovation and regional regulatory support, but initial results suggest Mistral’s model-centric focus and sparse architecture breakthroughs can proliferate LLMs democratically beyond Silicon Valley.

METHODOLOGY:

To write the following blog entry, CLAUDE 2, an LLM by Anthropic was used. Due to CLAUDE 2 lack of access to internet and the fact that, Claude was trained on data up until December 2022, it was unaware of the latest development in the field of Artificial Intelligence, or even about the existence of Mistral AI.

Due to that, I compiled a file based on the latest developments, articles and technical information available on the various models from Mistral AI and Open AI, and uploaded those to CLAUDE’S 2 context.

Following sources were used to upload into the CLAUDE 2:

https://spynewsletter.com/company/mistralai/

https://platform.openai.com/docs/models/model-endpoint-compatibility

https://medium.com/version-1/exploring-the-capabilities-of-gpt-4-turbo-d90d26df7174#:~:text=GPT%2D4%20Turbo%20boasts%20a,previous%20capacity%20of%2032%2C000%20tokens.

https://aibusiness.com/nlp/mistral-ai-s-new-language-model-aims-for-open-source-supremacy

https://www.pymnts.com/artificial-intelligence-2/2023/nvidia-invests-in-35-ai-companies-in-2023/#:~:text=Recently%2C%20Nvidia%20made%20a%20substantial,software%2C%20according%20to%20the%20report.

https://www.ft.com/content/293633cd-8a4c-4a7d-b14d-62a8a8b6c60a

https://www.ft.com/content/be680102-5543-4867-9996-6fc071cb9212

https://www.ft.com/content/25337df3-5b98-4dd1-b7a9-035dcc130d6a

https://www.ft.com/content/7e45b9a6-1f94-4229-b985-09958503b410

https://www.ft.com/content/9a06ddb4-b5c0-406c-b397-5684ba999c4d

https://www.ft.com/content/22c2aab0-74ed-4a36-933a-b30245275dea

https://www.ft.com/content/045878a7-a75a-47b9-bc7f-115ea1025c5b

https://www.reuters.com/technology/nvidia-may-be-forced-shift-out-some-countries-after-new-us-export-curbs-2023-10-17/

https://finance.yahoo.com/news/nvidia-sells-graphics-ai-chips-094500631.html

https://spynewsletter.com/company/mistralai/

https://platform.openai.com/docs/models/model-endpoint-compatibility

https://medium.com/version-1/exploring-the-capabilities-of-gpt-4-turbo-d90d26df7174#:~:text=GPT%2D4%20Turbo%20boasts%20a,previous%20capacity%20of%2032%2C000%20tokens.

https://aibusiness.com/nlp/mistral-ai-s-new-language-model-aims-for-open-source-supremacy

https://www.pymnts.com/artificial-intelligence-2/2023/nvidia-invests-in-35-ai-companies-in-2023/#:~:text=Recently%2C%20Nvidia%20made%20a%20substantial,software%2C%20according%20to%20the%20report.

https://www.ft.com/content/293633cd-8a4c-4a7d-b14d-62a8a8b6c60a

https://www.ft.com/content/be680102-5543-4867-9996-6fc071cb9212

https://www.ft.com/content/25337df3-5b98-4dd1-b7a9-035dcc130d6a

https://www.ft.com/content/7e45b9a6-1f94-4229-b985-09958503b410

https://www.ft.com/content/9a06ddb4-b5c0-406c-b397-5684ba999c4d

https://www.ft.com/content/22c2aab0-74ed-4a36-933a-b30245275dea

https://www.ft.com/content/045878a7-a75a-47b9-bc7f-115ea1025c5b

https://www.reuters.com/technology/nvidia-may-be-forced-shift-out-some-countries-after-new-us-export-curbs-2023-10-17/

https://finance.yahoo.com/news/nvidia-sells-graphics-ai-chips-094500631.html

Following prompts were used:

Write an academic paper comparing CHAT GPT LLM and MISTRAL AI LLM, and draw a conclusion to answer the question, “Will Mistral be Europe’s answer to Silicone Valleys LLMs?”

Nvidia has recently invested and partnered with MISTRAL AI, based on NVIDA dominance on the chips market used for AI, will this have an impact on the competition between OPEN AI and MISTRAL AI?

Will Mistral be Europe’s answer to Silicone Valleys LLMs?

Will Mistral be Europe’s answer to Silicone Valleys LLMs? Answer with the thesis, that “Mistral won’t be Europe’s answer to Silicone Valleys LLMs.”

Reading Time: 3 minutes

On Friday, 22 of December 2023, the head of Allen & Overy Luxembourg, Patrick Mischo, presented ` new artificial intelligence-powered tool that 1,000 of the firm’s lawyers are already using. ContractMatrix is an AI-powered contract drafting tool that helps users create legally sound contracts, in a fraction of the time it would take to do so manually. It was developed by Allen & Overy, a British multinational law firm headquartered in London, in partnership with Microsoft and legal AI start-up Harvey. It’s a powerful tool that can help businesses of all sizes create legally sound contracts, in a more efficient and cost-effective manner. 

In a trial run, Dutch chipmaking equipment manufacturer ASML and health technology company Philips said they used the service to negotiate what they called the “world’s first 100 per cent AI generated contract between two companies”.

How does it work?

ContractMatrix uses a combination of natural language processing (NLP), machine learning, and legal expertise to generate contracts that are tailored to the specific needs of the user. The tool can also be used to review existing contracts and identify potential issues.

Key features of ContractMatrix:

• Automated contract drafting: ContractMatrix can generate contracts for a wide variety of use cases, including employment contracts, non-disclosure agreements, and purchase orders.

• Legally sound contracts: ContractMatrix is powered by a team of experienced lawyers who ensure that all contracts are compliant with relevant laws and regulations.

• Customizable templates: ContractMatrix offers a library of pre-built templates that can be customized to meet the specific needs of the user.

• Real-time feedback: ContractMatrix provides real-time feedback on the user’s contract, highlighting potential issues and suggesting improvements.

• Easy to use: ContractMatrix is designed to be easy to use, even for those who are not familiar with legal terminology.

Benefits of using ContractMatrix:

• Save time: ContractMatrix can save users a significant amount of time by automating the drafting process.

• Reduce errors: ContractMatrix can help to reduce the risk of errors in contracts by providing real-time feedback.

• Improve legal compliance: ContractMatrix can help to ensure that contracts are compliant with relevant laws and regulations.
• Enhance legal expertise: ContractMatrix can provide users with access to legal expertise without the need to hire a lawyer.

• Reduced cost: The use of ContractMatrix can save businesses money by reducing the time and cost of contract negotiation.

• Reduced cost: The use of ContractMatrix can save businesses money by reducing the time and cost of contract negotiation.

Concerns about the use of ContractMatrix:

• Accuracy and Legality: ContractMatrix relies on its algorithms and training data to generate contracts, which may lead to inaccuracies or errors in the legal language, which could have significant consequences for the parties involved in the contract.

• Compliance and Updatability: ContractMatrix needs to be continuously updated with new legal precedents, regulations, and evolving industry standards to ensure that generated contracts remain compliant and up to date. 

• Ethical Considerations: AI-powered tools like ContractMatrix raise ethical concerns, such as the potential for discrimination or bias in the generation of contracts based on factors like race, gender, or socioeconomic status.

• Data Security and Privacy: ContractMatrix collects and processes sensitive data, such as personal information and business details, making it crucial to implement robust data security measures to protect user privacy and prevent unauthorized access or breaches.

• User Training and Education: Users should receive adequate training and education on how to effectively utilize ContractMatrix, including understanding its limitations, interpreting its feedback, and making informed decisions about the generated contracts.

Impact of ContractMatrix:

Is ContractMatrix indeed the future of contract negotiation? I would argue yes. It soon will be a necessary tool for any large business or lawyer to have, but it will still require trained personnel, with the ability and knowledge to use it properly and check its work. While it can automate many of the time-consuming and repetitive tasks involved in contract negotiation, such as: identifying boilerplate language, identifying potential risks and issues, and suggesting alternative language, its only purpose is to cut the workload and time that the process of contract negotiating take, it will not be able to take lawyers place in the foreseeable future. But it will save lawyers a significant amount of time and effort, allowing them to focus on more complex and strategic aspects of the negotiation process. Furthermore, it will have a large financial impact for the big law, while currently in many contracts’ negotiations, corporations inhouse lawyers simply aren’t able to deal with the workload and require outside help with different projects, ContractMatrix will allow them to complete their tasks without outside help in a significantly faster manner. It will also have a positive impact on businesses themselves, smaller legal costs will allow them to invest in other areas and expand faster. 

Sources:

https://www.allenovery.com/en-gb/global/expertise/advanced_delivery/contractmatrix

https://www.ft.com/content/f1aff4d0-b2c5-4266-aa0a-604ef14894bb

https://delano.lu/article/contractmatrix-allen-overy-s-n

https://medium.com/@multiplatform.ai/law-firm-allen-overy-unveils-ai-contract-negotiation-tool-fc83b155bb29

https://www.capitalbrief.com/briefing/allen-overy-launches-ai-contract-drafting-tool-6a64c1ec-7aa8-41ad-8fc7-81f4c40f087d/

https://bard.google.com/chat

Bard queries:

How does ContractMatrix, new AI-based contract drafting tool work?

Who created ContractMatrix?

Are there any concerns regarding the usage of ContractMatrix?

Is ContractMatrix indeed the future of contract negotiation.

Reading Time: 6 minutes

Recently the Pentagon has announced the “Replicator” initiative, which first task will be to quickly scale and field thousands of attritable autonomous systems within the next 18 to 24 months, leveraging AI, robotics, and commercial technology. But this is just the beginning, AI has the potential to revolutionize warfare in many ways. The following article will examine those ways with focus on three distinct aspects of the use of AI in warfare: cyberwarfare, gathering, processing information and decision making and autonomous warfare. 

Cyberwarfare

Current Uses of AI in Cyberwarfare:

1. Threat Detection and Mitigation: One of the primary uses of AI in cyberwarfare is threat detection and mitigation. AI algorithms are trained to analyze large amounts of data and identify patterns and anomalies that could indicate a cyber-attack. This allows for early detection and response to potential threats, thereby reducing the risk of a successful attack.

2. Automated Malware Detection and Removal: AI-powered antivirus software can automatically detect and remove malware from systems, without the need for human intervention. This not only saves time and resources, but also improves the effectiveness of malware detection and removal.

3. Cyberattack Planning and Execution: AI can also be used to plan and execute cyber-attacks. By analyzing the vulnerabilities of a target system, AI algorithms can identify the best approach for a successful attack and automatically launch it. This reduces the need for human involvement and can make cyber-attacks more efficient and effective.

4. Social Engineering Attacks: Social engineering attacks, such as phishing, are a common tactic used by cyber criminals to gain access to sensitive information. AI-powered tools can analyze social media and other online data to create personalized and convincing phishing messages, making it easier to trick individuals into giving away their personal information.

Future Uses of AI in Cyberwarfare:

1. Autonomous Cyber Weapons: With the advancement of AI, it is possible that we could see the development of autonomous cyber weapons. These weapons would be able to identify and attack targets without any human intervention, making them more difficult to defend against.

2. Predictive Analytics: AI can also be used for predictive analytics in cyberwarfare. By analyzing past cyber-attacks and their outcomes, AI algorithms can predict future attack patterns and help organizations prepare for potential threats.

3. Cyber Defense Strategies: AI can help in developing more effective cyber defense strategies by continuously monitoring and analyzing data to identify vulnerabilities and potential attack patterns. This can help organizations stay one step ahead of cybercriminals.

4. Cyber Intelligence: AI can also be used to gather intelligence on potential cyber threats. By analyzing data from various sources, including social media and the dark web, AI algorithms can identify potential threats and provide insights on the motives and methods of cyber attackers.

Gathering, processing information and decision making

Current Uses of AI 

One of the most prominent uses of AI in warfare decision making is in the field of autonomous weapons systems. These are weapons that can identify and engage targets without direct human involvement. The use of autonomous weapons has been controversial, with concerns about the potential for these systems to malfunction or be used for unethical purposes. However, proponents argue that these systems can make quicker and more accurate decisions than humans, reducing the risk of human error and casualties.

AI is also being used in intelligence gathering and analysis. With the vast amount of data collected in modern warfare, AI can quickly sift through and analyze this data to identify patterns and insights that may not be visible to human analysts. This can provide commanders with valuable information to make informed decisions on the battlefield.

Another area where AI is being utilized is in logistics and supply chain management. With the use of predictive analytics, AI can forecast future demands and optimize the allocation of resources, leading to more efficient and effective operations. This is especially crucial in times of crisis or in remote and hostile environments where supply chains may be disrupted.

Future Uses of AI 

As technology continues to advance, the use of AI in warfare decision making is expected to expand even further. 

AI is also being explored for use in decision-making support for military commanders. By analyzing data from various sources, AI can provide commanders with real-time situational awareness and aid in identifying potential threats and opportunities. This can help commanders make more informed and timely decisions on the battlefield.

In addition to decision-making support, AI is also being researched for use in autonomous decision making. This would involve AI systems having the ability to make critical decisions without human intervention. While this may still be in its early stages, the potential for AI to make decisions at a speed and accuracy that surpasses human capabilities is a promising prospect for the future of warfare.

Autonomous warfare

Current Uses of AI in Warfare:

AI has been used in warfare for decades, primarily in the form of drones. These unmanned aerial vehicles (UAVs) use AI algorithms to navigate and collect data in real-time, making them valuable assets in intelligence, surveillance, and reconnaissance (ISR) missions. AI also plays a critical role in target identification and precision strikes, making military operations more efficient and reducing the risk of collateral damage.

Future Uses of AI in Warfare:

The future uses of AI in warfare are vast and diverse, with the potential to revolutionize the way wars are fought. One of the most anticipated developments is the use of autonomous weapons, also known as lethal autonomous weapons systems (LAWS). These weapons can operate without human intervention and have the ability to make decisions and engage targets independently. This technology has the potential to reduce the number of casualties on the battlefield and increase the speed and accuracy of military operations.

Implications of Autonomous Weapons:

While the use of autonomous weapons may seem like a promising advancement in warfare, it also raises significant ethical concerns. One of the primary concerns is the lack of human control and accountability in the decision-making process. The use of autonomous weapons could potentially lead to unintended consequences and the loss of innocent lives. Additionally, there are concerns about the potential for these weapons to be hacked and used against their creators, as well as the possibility of a global arms race for AI-powered weapons.

Another concern is the moral and legal implications of delegating life and death decisions to machines. The use of autonomous weapons raises questions about the responsibility and accountability of those involved in their development and deployment. It also challenges the principles of international humanitarian law, which requires human judgment and control in the use of weapons.

Conclusion:

The use of AI in warfare has already proven to be a game-changer, with its ability to enhance military capabilities and decision-making processes. However, the further incorporation of AI, particularly in the form of autonomous weapons and delegating the decision making to AI, raises significant ethical concerns and requires careful consideration and regulation. It is crucial for governments and international organizations to have open and transparent discussions about the use of AI in warfare and establish guidelines to ensure the responsible and ethical development and deployment of this technology.

Yet, in the view of this author, in foreseeable largescale conflicts, such as in a hypothetical conventional war between USA and China, the AI willy only hold a backseat. The bulk of decision making, and the actual fighting, will still be by humans and between humans. This sentiment has been somehow proven in the Russo-Ukrainian War, in which despite the use of new technology and equipment and the use of autonomous drones, both sides have continued  relaying on infantry, as the main fighting force. Technology, even backed by AI, is simply less dependable, more prone to malfunctions and far more vulnerable to electronic warfare and cyberattacks, then the human component, and most importantly, often far more expensive and prone to being disrupted by logistical issues.  While the initial clash, just as was the case in the Russo-Ukrainian war, maybe the clash between the technology, with the use of the newest state of art equipment, as long as the sides will be somehow evenly matched, it’s hard to imagine AI being the deciding factor. Every measure, has a counter measure, and the characteristics of the AI, it’s lack of creativity, ecosystems based on cloud technology, reliance on historical date, poor adaptability, make it significantly easier for those countermeasures to be developed and employed. 

. 

Sources:

https://www.hudson.org/defense-strategy/artificial-intelligence-future-warfare

AI and the future of warfare: The troubling evidence from the US military

https://www.cbsnews.com/news/artificial-intelligence-in-military-general-mark-milley-future-of-warfare-60-minutes/

https://www.cna.org/reports/2023/10/ai-and-autonomous-technologies-in-the-war-in-ukraine

Ukraine’s Secret Weapon – Artificial Intelligence

AI:

In writing the above blog entry, toolbaz.com generative AI was used. Following prompts were used:

• Analyze the current and future uses of AI in cyberwarfare.
• Analyze the current and future uses of AI in cyberwarfare, in a form of a blog entry.
• Analyze the current and future uses of AI in warfare decision making, in a form of a blog entry.

Reading Time: 4 minutes

The AI act is a proposed regulation that aims to establish harmonized rules on artificial intelligence in the EU. It is based on a risk-based approach, meaning that the higher the risk of an AI system, the stricter the rules it has to comply with. The AI act also intends to foster innovation and investment in AI across Europe, while ensuring that AI systems respect fundamental rights and values.

The impact of the AI act on the European artificial industry could be significant, as it would create a common legal framework and a level playing field for AI providers and users in the EU. 

The proposed AI Act (European Commission, 2021), which at time of writing is still to be finalised between the EU institutions 2 , seems to be a poor fit for foundation models. It follows the risk-approach, following the idea, that each AI application can be put into a specific risk category, taking into account it’s intended application. This follows the traditional EU approach, where a single product can be neatly pute into a single category, but this ignores the specifics of the AI technology and foundation models, which easily can be customized to great many potential uses. 

In the ongoing legislative work to amend the text, the European Parliament has proposed that providers of foundation models perform basic due diligence on their offerings. In particular, this should include:

• Risk identification. Even though it is not possible to identify in advance all potential use cases of a foundation model, providers are typically aware of certain vectors of risk. OpenAI knew, for instance, that the training dataset for GPT-4 featured certain language biases because over 60 percent of all websites are in English. The European Parliament would make it mandatory to identify and mitigate reasonably foreseeable risks, in this case inaccuracy and discrimination, with the support of independent experts.
• Testing. Providers should seek to ensure that foundation models achieve appropriate levels of performance, predictability, interpretability, safety and cybersecurity. Since the foundation model functions as a building block for many downstream AI systems, it should meet certain minimum standards.
• Documentation. Providers of foundation models would be required to provide substantial documentation and intelligible usage instructions. This is essential not only to help downstream AI system providers better understand what exactly they are refining or fine-tuning, but also to enable them to comply with any regulatory requirements.

The impact of the AI act on the European artificial industry could be quite significant.

Yet, some of the potential benefits of the AI act for the industry are:

• Increased trust and acceptance of AI by consumers and society, as they can be assured that AI systems are safe, reliable and ethical.
• Reduced fragmentation and legal uncertainty, as the AI act would harmonise the rules and requirements for AI across the EU, and provide a single market for AI products and services.
• Enhanced competitiveness and innovation, as the AI act would stimulate research and development on AI, support the uptake of AI by small and medium-sized enterprises, and facilitate the access to data and computing resources.
• Greater influence and leadership, as the AI act would set a global standard for AI regulation, and promote the European approach to AI in the international arena.

However, the AI act could also pose some challenges and costs for the industry, such as:

• Increased compliance burden and administrative costs, as the AI act would impose new obligations and responsibilities for AI providers and users, especially for high-risk AI systems.
• Potential trade barriers and market distortions, as the AI act could create differences and conflicts with other jurisdictions that have different or no rules on AI, and affect the competitiveness of European AI actors in the global market.
• Possible unintended consequences and risks, as the AI act could have negative impacts on innovation, diversity, and inclusion, if it is not implemented and enforced in a balanced and proportionate way.

The French president Emanuel Macron, has repeatedly warned against stifling innovation with the AI act. It’s hard to see him as objective though, with the large sums invested by the French government into the development of AI. 

Proposals to regulate AI are a source of concern for France and the French companies involved, insofar as some feel that they may stifle innovation by imposing overly burdensome obligations.

Thierry Breton, the French Commissioner for the Internal Market of the European Union, has defended the proposed new regulatory scheme against threats of some CEOs in the technology sector that they would leave the EU market because of the rigidity of the proposal: “No companies will do that, as the EU market is the biggest digital market, and we welcome everyone.”

Personally, I believe, that the AI act is something that is extremely necessary to provide a legal framework for the uses of the Artificial intelligence. The general text of the AI act will be only seen as guidelines to be followed, and the practice in various countries will be the determining factor, whether it will stifle or encourage innovation in the field. As of today, we still do know what the final version of the AI act will be. Furthermore, it will become law years from today, giving everyone ample time necessary to prepare for the new regulations, while similarly protecting us from a scenario in which the AI field will be the new wild west. 

AI used – BING CHAT:

• What will the impact of AI act be on the European artificial intelligence industry?

Sources:

https://economictimes.indiatimes.com/tech/technology/frances-macron-warns-against-punitive-ai-regulation/articleshow/105297612.cms?from=mdr

https://www.kramerlevin.com/en/perspectives-search/frances-take-on-artificial-intelligence-and-on-the-eu-artificial-intelligence-act-under-discussion.html

https://www.bruegel.org/analysis/adapting-european-union-ai-act-deal-generative-artificial-intelligence

https://www.ey.com/en_ch/forensic-integrity-services/the-eu-ai-act-what-it-means-for-your-business#

https://www.europarl.europa.eu/news/en/headlines/society/20230601STO93804/eu-ai-act-first-regulation-on-artificial-intelligence

Reading Time: 4 minutes

Since purchasing Twitter in October last year for the eye watering sum of 44 billion dollars, Elon Musk has fired over 80% of the Twitter workforce, rebranded the company to X, lost over 50% of the ad revenue and promoted headlines such as “Elon Musk Really Broke Twitter This Time” (The Atlantic) ,“Elon Musk’s Twitter is dying a slow and tedious death” (Financial Times) or “Musk is nearly done destroying what made Twitter Twitter” (The Washington Post).

But is X really dying? And did Musk kill Twitter?  Even when ignoring the disappearance of the iconic bird, changes to X have been significant in the past year. 

1. Changes to the verification system with the launch of Twitter Blue, a subscription service that gives users a blue verification checkmark. In May 2023 Musk has announced that Twitter would no longer verify accounts based on their public interest.
2. Layoffs and staff turnover. In April 2023, Musk has stated that 80% of the workforce since he took over has been fired. He has also seen a high rate of staff turnover, as many employees have left the company due to Musk’s controversial policies and management style
3. Relaxed content moderation policies, in an attempt to make X a more open and free platform.
4. Increased focus on algorithmic transparency. Musk has said that he wants to make Twitter’s algorithms more transparent. In February 2023, Twitter launched a new tool that allows users to see how their tweets are being ranked in the algorithm.
5. Changes to the API. Musk has also made changes to Twitter’s API, which is used by third-party developers to create apps and services that interact with Twitter. These changes have made it more difficult for developers to build and maintain Twitter apps.
6. Increased promotion of Musk’s other businesses. Musk has used Twitter to promote his other businesses, such as Tesla and SpaceX. This has led to accusations that Musk is using Twitter to benefit his own interests.

The results have been easy to predict. 

1. Decline in user satisfaction. Many users are unhappy with Musk’s changes to the platform, such as the relaxed content moderation policies and the increased promotion of Musk’s other businesses.
2. Increase in hate speech. According to a study by the Center for Countering Digital Hate, the amount of hate speech on Twitter has increased by 50% since Musk acquired the platform.
3. Decline in number of active users. According to Statista, the number of active users on Twitter has been steadily declining since Musk acquired the platform in October 2022. In the fourth quarter of 2022, Twitter had 217 million active users. In the first quarter of 2023, that number had fallen to 211 million. 
4. Increased misinformation. According to a study by Stanford University, the amount of misinformation on Twitter has increased by 25% since Musk acquired the platform. 
5. Drop in ad revenue. According to Elon Musk himself, Twitter has lost roughly half of its advertising revenue since he bought the company in October 2022. 

Yet, the layoffs and others cost savings cuts made by Musk have saved the company billions of dollars. According to a report in the Financial Times, Elon Musk has cut $4.5 billion in costs at Twitter as of October 2023. Simultaneously, Twitter lost “only” $2.23 Billion in Ad revenue since Musk takeover. In a grand scheme of things, the company might actually for the better with the changes. Obviously we cannot simply discount the increased number of misinformation, hate speech and drop in user satisfaction and user numbers – in the long term those might prove to be the undoing of X – but it would be naive to think that nor Musk, nor any of his advisors weren’t expecting those effects when firing 80% of the workforce. It seems that X can only improve in those areas now, which in turn, will attract the return of more advertisers.

It’s also important to note, that Musk didn’t exactly buy Twitter for Twitter. He bought it for the hundreds of millions of users in an attempt to create the “everything app”, as he tweeted even before acquiring the social media giant. With a project of such scale, drastic changes were to be expected.

While, it is still too early to say what the long-term impact of Musk’s ownership of X will be, if one were to anthropomorphize X, one could expect it to quote Mark Twain and state “The report of my death was an exaggeration”.  

Musk has killed Twitter in a shape that we knew, yet X might be here to stay.

By Szymon Olszewski

Sources:

• https://bard.google.com/chat

Prompts used:

1. Did Elon Musk kill twitter? Provide statistical data to back up your claim.
2. How did Twitter change since Elon Musk bought the company in October 2022? Provide 10 examples.
3. How much costs have Musk cut since he bought Twitter. Provide exact numbers with sources.
4. How much total cost cuts have been made by Musk to Twitter as of October 2023. Provide sources.
5. How much advertising revenue has Twitter lost as of October 2023, since Musk bought Twitter. Use Financial Times and provide the title of your source article.

• https://www.washingtonpost.com/politics/2023/10/05/musk-twitter-news/

• https://www.ft.com/content/8180fb22-60ed-4df1-8f62-3752fae71dc1

• https://www.ft.com/content/b429b624-bf82-4ccd-bf69-b75055403952

• https://www.businesstoday.in/technology/news/story/elon-musk-confirms-he-has-fired-over-80-of-twitter-employees-so-far-377045-2023-04-12

• https://www.ft.com/content/a149046c-5409-4390-9cf0-bd4c2b17c251

• https://www.theguardian.com/media/2023/jul/29/elon-musk-wechat-twitter-rebranding-everything-app-for-west

• https://www.demandsage.com/wechat-statistics/