37signals' Operations team is moving their applications off of the cloud and back onto their own hardware due to technical debt and complexity of cloud services.
They created a tool called 'mrsk' to simplify the operation of their existing containerized applications.
They modernized and simplified their configuration management to cut the time for provisioning virtual machines from about 20 minutes per guest to less than a minute.
37signals switched to Cloudflare for their CDN and DNS requirements to replace CloudFront and Route53.
Tatalist, Writeboard, and Backpack were migrated in less than six weeks with the last migration taking only 16 minutes.
Most of their apps only required a standard Rails app deployment as infrastructure requirements.
Sales of previously owned homes in the US rose 14.5% in February 2019 from the prior month, but were down 22.6% from a year earlier.
Home prices in the US fell in February for the first time in 11 years, showing the effect of the Federal Reserve's campaign to raise interest rates.
The National Association of Realtors reported that the US experienced the first year-over-year drop in home prices in more than a decade and a dip in mortgage rates.
OpenAI's policies are hindering reproducible research on language models, which have become privately-controlled research infrastructure.
OpenAI's model deprecations are a concern because Language Models (LLMs) are becoming key pieces of infrastructure, and researchers and developers rely on LLMs as a foundation layer, which is then fine-tuned for specific applications or answering research questions.
OpenAI, like other companies, regularly updates newer models, such as GPT-3.5 and GPT-4, meaning the use of those models can be a barrier to reproducibility.
OpenAI's hasty deprecation of Codex has forced researchers and developers to switch to GPT 3.5 models, which are not comparable, leading to the irreproducibility of old work.
OpenAI has allowed researchers access to Codex, but the application process is opaque, and it's unclear how long the model will be available.
Open-sourcing LLMs could be a key step in ensuring reproducibility.
Mozilla has announced the creation of Mozilla.ai, which will build a trustworthy and independent open-source AI ecosystem.
The startup will make it easy to develop trustworthy AI products with agency, accountability, transparency, and openness at its core.
Mozilla is committing $30 million to this initiative and will be led by Moez Draief.
The initial focus will be on tools that make generative AI safer and more transparent, as well as people-centric recommend systems that do not misinform or undermine well-being.
The Securities and Exchange Commission (SEC) has charged crypto asset entrepreneur Justin Sun and three of his companies for the unregistered offer and sale of crypto asset securities Tronix (TRX) and BitTorrent (BTT).
The SEC alleges Sun and his companies offered and sold TRX and BTT as investments through multiple unregistered "bounty programs."
Sun also violated the antifraud and market manipulation provisions of the federal securities laws by orchestrating a scheme to artificially inflate the apparent trading volume of TRX in the secondary market.
The SEC also charged eight celebrities for illegally touting TRX and/or BTT without disclosing compensation: Lindsay Lohan, Jake Paul, DeAndre Cortez Way (Soulja Boy), Austin Mahone, Michele Mason (Kendra Lust), Miles Parks McCollum (Lil Yachty), Shaffer Smith (Ne-Yo), and Aliaune Thiam (Akon).
With the exception of two celebrities, the charged individuals agreed to pay a total of more than $400,000 in disgorgement, interest, and penalties to settle the charges, without admitting or denying the SEC's findings.
A blog post argues against using Raspberry Pi as the default choice for unobtrusive and inexpensive Linux hosts.
The author suggests that Raspberry Pi should only be used when it excels in specific needs like GPIO interfaces, extremely low power requirements, or tight space constraints.
The author provides examples of how other options such as AMD GX-415GA or HP Prodesk 600 G4 micro desktops could provide more computing power, faster and more reliable storage, and better community support for similar or even lower costs.
The author concludes that using a more powerful system that can cover a wide range of purposes is better in the long run compared to accumulating a bunch of limited-purpose Raspberry Pi's.
The developer job market is currently highly competitive, making it difficult for candidates to find suitable roles.
Some companies are using excessively lengthy hiring processes that include multiple interviews, personality quizzes, and technical tests, which are putting potentially hireable candidates off.
Negative experiences regarding hiring practices can be reported through online reviews as a way of raising awareness about problems with recruitment methods.
Candidates find job interviews to be a pass/fail event, whereas employers view them as a competition.
Feedback is rarely given as companies have policies to avoid possible discrimination lawsuits.
Candidates should not waste time applying for jobs that are not a good match and LinkedIn plays a key role in helping candidates to track applications and communicate with recruiters.
Some companies have over-complicated interview processes for jobs that are simple.
The job market for software engineers is tough right now, with many high-quality engineers on the market but companies being cautious and choosy in their hiring.
Companies are being more selective in their hiring process, resulting in longer interview processes and more competition for roles.
Networking remains the best way to find jobs, particularly in a tough market.
A study by Stanford researchers reveals that people can only accurately identify AI-generated text about 50% of the time.
Participants in the study were presented with text samples across social media platforms of OKCupid, AirBNB, and Guru.com.
The study found that high grammatical correctness and the use of first-person pronouns were often attributed incorrectly to human-generated text, as were references to family life and the use of informal, conversational language.
The implications of the study suggest that cheaper and easier AI-generated content could lead to more misinformation in the future.
The research team suggests solutions such as giving AI a recognizable accent or self-disclosing machines in high-stakes scenarios to improve our ability to differentiate between human and AI-generated content.