July broadens the choices facing business leaders. Downloadable models, longer-context assistants, and a work-oriented search chat are not interchangeable purchases. Each changes where information goes, who runs the system, and how the work is checked. The right decision begins with the problem the business needs to solve. A smaller, well-managed deployment may be more useful than a more ambitious system that the organization cannot support or evaluate consistently.
1. Llama 2 expands the model deployment choices
Meta and Microsoft announced Llama 2 on July 18, making the model available for research and commercial use under its license. Meta described distribution through partners as well as downloadable access. The release gives developers another foundation for applications; it does not make the operational work of running and securing those applications disappear.
A company considering its own deployment should distinguish access to model weights from having a complete business service. Infrastructure, updates, access controls, monitoring, and employee support still need owners. A managed service may suit one firm while a more customized deployment suits another. The comparison depends on requirements and capacity, not on a simple free-versus-paid label.
Business takeaway: Compare total operating responsibility before choosing a route. List who will maintain the environment, test changes, respond to failures, and review the model's license conditions. Use the same business tasks to evaluate hosted and self-managed options. Treat greater control as a responsibility to plan for, rather than assuming it automatically produces lower cost or better results.
Source: https://about.fb.com/news/2023/07/llama-2/
2. Claude 2 brings longer inputs into the discussion
Anthropic introduced Claude 2 on July 11 through its API and a public beta chat experience in the United States and United Kingdom. The company described support for inputs up to 100,000 tokens and improvements in coding, mathematics, and reasoning. These are provider-reported capabilities; a large input limit does not establish perfect recall or correct interpretation of every passage.
For a business reviewing long internal manuals, the possibility of working with more material at once is attractive. But a summary can omit a narrow exception that matters to the decision. Employees should still be able to trace an important conclusion to its source and identify conflicting instructions in the supplied material.
Business takeaway: Evaluate a document assistant with questions whose answers appear in different parts of a known document set. Include one question where the answer is absent and one where two versions disagree. Require references to the relevant passages. Measure how often the employee can verify the answer quickly, rather than judging the system only by how much text it can accept.
Source: https://www.anthropic.com/news/claude-2
3. Bing Chat Enterprise adds work-oriented data commitments
Microsoft announced Bing Chat Enterprise in preview on July 18. The company described commercial data protection, including not using chat data to train models and not retaining chat data. In the same announcement, Microsoft disclosed planned Microsoft 365 Copilot pricing for broader availability later. The preview and the pricing announcement should not be mistaken for general availability of every Copilot feature.
For an employer, the useful distinction is between a consumer account and a business arrangement with specific data commitments. Employees need a clear approved route, and administrators need to understand which account and settings actually receive those protections. A label alone is not a complete deployment policy.
Business takeaway: Before allowing work information into an assistant, verify the account type, applicable terms, and access configuration. Give staff an approved entry point and a short explanation of permitted use. Test whether source links support generated research summaries. Data commitments help address one concern, but they do not remove the need to check the content of an answer.
The management decision
A useful selection meeting should end with a test, not a preferred brand. Choose a workflow, define the information it needs, and compare alternatives against the same acceptance criteria. Include the ongoing work of administration and review in the estimate. A technically capable option can still be a poor fit if the business lacks the time or staff to operate it. Keep the initial scope small enough that a disappointing result can be corrected without disrupting customers or essential internal processes.
Present-day invitation
Want to turn these ideas into a practical improvement for your business today? Book an appointment with Brice Prather at https://loyti.com/book to discuss AI, custom software, or business automation. Bring a question, a costly bottleneck, or a growth opportunity. Together, we can identify a sensible next step, what it would require, and how its value should be measured.